Escolar Documentos
Profissional Documentos
Cultura Documentos
Júri
Presidente: Professor Doutor João Manuel Lage de Miranda Lemos
Orientador: Professor Doutor Nuno Manuel Mendes Maia
Co-Orientador: Professor Doutor Fernando José Parracho Lau
Vogais: Professor Doutor Filipe Szolnoky Ramos Pinto Cunha
Professor Doutor Miguel António Lopes de Matos Neves
Outubro 2011
ACKNOWLEDGEMENTS
This work was the result of intense work developed in the last 6 months. The effort was turned
much lighter with the cooperation carried out with my college Rafael Carvalho and the PhD student
Yoann Lage, who are developing projects in this subject, and have supplied me with valuable
information relevant to the content of this work.
I must also leave here a word of gratitude to Professor Miguel Neves for his support and
revision of the numeric methods employed during the course of this work. My thanks should also be
addressed to my advisers Professor Nuno Maia and Professor Fernando Lau. The first for is revision
and support in the area of the vibration theory and the second also for having reviewed this work and
for his advices concerning the Propfan engine case-study.
Finally I must thank my family to whom I owe it all.
i
RESUMO
Palavras-chave:
Dinâmica de Rotors, Identificação de Forças, Domínio do Tempo, Propfan, SWAT, ISF,
DMISF
ii
ABSTRACT
Keywords:
Rotordynamics, Force Identification, Time-domain, Propfan, SWAT, ISF, DMISF
iii
TABLE OF CONTENTS
LIST OF FIGURES .................................................................................................................................. vi
NOMENCLATURE .................................................................................................................................. ix
1 INTRODUCTION ............................................................................................................................. 1
2 FUNDAMENTALS ........................................................................................................................... 6
iv
3.2.5 DMISF ................................................................................................................................ 45
REFERENCES ...................................................................................................................................... 84
APPENDIXES ........................................................................................................................................ 86
Appendix 4 – Deviation Between the Natural Frequencies Obtained via Direct Method and the
Pseudo-modal Method .......................................................................................................................... 96
v
LIST OF FIGURES
Figure 2.1 – Amplitude evolution with time on an unstable structural system subject to an initial
disturbance – source: (7) ......................................................................................................................... 7
Figure 2.2 – The disk and its reference frames – source: (4) ................................................................. 8
Figure 2.3 – The cross-section of the shaft – source: (4) ........................................................................ 9
Figure 2.4 – The bearing model – source: (4) ....................................................................................... 10
Figure 2.5 – Rotating mass with offset length – source: (4) .................................................................. 11
Figure 2.6 – The Multirotor model ......................................................................................................... 12
Figure 2.7 – Inertial reference frame coordinates – source: (4) ............................................................ 13
Figure 2.8 – Campbell Diagram example – source: (4) ........................................................................ 17
Figure 2.9 – Response diagram example – source: (4) ........................................................................ 19
Figure 2.10 – Possible orbits of a rotor’s shaft ...................................................................................... 20
Figure 2.11 – Generic elliptical orbit with displacement vector ..................................................... 21
Figure 2.12 – Position of a mass unbalance (in bold) in a disk performing Forward and Backward Whirl
............................................................................................................................................................... 22
Figure 2.13 – Illustration of a decomposition of a time-domain spectrum – source: (7) ....................... 25
Figure 3.1 – Shaft finite element – source: (4) ...................................................................................... 37
Figure 3.2 – Force components due to the presence of a mass unbalance in the disk of the rotor. .... 46
Figure 4.1 – The Multirotor model ......................................................................................................... 48
Figure 4.2 – Campbell Diagram convergence – the brackets above represent: (number of elements in
the inner shaft/number of elements in the outer shaft) .......................................................................... 50
Figure 4.3 – Campbell diagram – Symmetric rotor Example 1 ............................................................. 51
Figure 4.4 – Finite element curves make use of the direct method (chapter 3.1.3) .............................. 52
Figure 4.5 – Finite element curves make use of the direct method (chapter 3.1.3) .............................. 53
Figure 4.6 – Campbell diagram – Symmetric rotor Example 2 ............................................................. 54
Figure 4.7 – Finite element curves make use of the direct method (chapter 3.1.3) .............................. 55
Figure 4.8 – Finite element curves make use of the direct method (chapter 3.1.3) .............................. 56
Figure 4.9 – Campbell diagram – Symmetric Multirotor Example 3 ...................................................... 56
Figure 4.10 – Finite element curves make use of the direct method (chapter 3.1.3) ............................ 57
Figure 4.11 – Finite element curves make use of the direct method (chapter 3.1.3) ............................ 58
Figure 4.12 – Campbell diagram – Asymmetric Multirotor .................................................................... 59
Figure 4.13 – Finite element curves make use of the direct method (chapter 3.1.3) ............................ 60
Figure 4.14 – Campbell Diagram – Damped Multirotor Example 1....................................................... 61
Figure 4.15 – Results make use of the direct method (chapter 3.1.3) .................................................. 61
Figure 4.16 – Campbell Diagram – Damped Multirotor Example 2....................................................... 62
Figure 4.17 – Results make use of the direct method (chapter 3.1.3) .................................................. 63
Figure 4.18 – Beam Model with Force Impulse position represented by vector F ................................ 64
Figure 4.19 – Finite Element Model of the Beam .................................................................................. 64
Figure 4.20 – SWAT applied to the beam ............................................................................................. 66
vi
Figure 4.21 – ISF applied to the beam .................................................................................................. 67
Figure 4.22 – DMISF applied to the beam ............................................................................................ 68
Figure 4.23 – The Symmetric rotor model ............................................................................................. 68
Figure 4.24 – Finite element model with the element numbers in black and the nodal numbers in blue
............................................................................................................................................................... 69
Figure 4.25 – Error vs Force Combination Graphic .............................................................................. 71
Figure 4.26 – SWAT applied to the Monorotor ...................................................................................... 72
Figure 4.27 – ISF applied to the Monorotor ........................................................................................... 72
Figure 4.28 – DMISF applied to the Monorotor ..................................................................................... 73
Figure 4.29 – The DUPRIN Propfan in its test facility – source: (6) ...................................................... 74
Figure 4.30 – The Propfan revolution section ....................................................................................... 75
Figure 4.31 – The Propfan finite element model ................................................................................... 76
Figure 4.32 – Propfan Campbell diagram comparison .......................................................................... 77
Figure 4.33 – First two modal shapes of the Propfan; Author’s model above/Model in (6) below; FW
(Left) and BW(Right) .............................................................................................................................. 78
Figure 4.34 – ISF applied to an accelerating Propfan ........................................................................... 80
Figure A.1 – Whirl sense and orbit phase with .............................................................. 87
Figure A.2 – Whirl sense and orbit phase with .............................................................. 88
Figure A.3 – Generic periodic orbit of a rotor relative to a inertial reference frame– source: (7) - edited
............................................................................................................................................................... 90
Figure A.4 – Force components of the various matrices decomposed on its symmetric and skew-
symmetric terms – source: (7) ............................................................................................................... 92
Figure A.5 – First bending mode shape of the Multirotor ...................................................................... 96
Figure A.6 .............................................................................................................................................. 97
Figure A.7 .............................................................................................................................................. 97
Figure A.8 – Beam Element and its Degrees of Freedom .................................................................... 98
vii
LIST OF TABLES
Table 4.1 – Critical speeds of each mesh and the correspondent errors computed using the FE
solution as reference ............................................................................................................................. 50
Table 4.2 – Critical speeds associated with the previous Campbell diagram – RR/FE error ............... 51
Table 4.3 – Critical speeds associated with the previous Campbell diagram – RR/FE error ............... 55
Table 4.4 – Critical speeds associated with the previous Campbell diagram – RR/FE error ............... 57
Table 4.5 – Critical speeds associated with the previous Campbell diagram – RR/FE error ............... 59
Table 4.6 – Critical speeds associated with the previous Campbell diagram ....................................... 61
Table 4.7 – Critical speeds associated with the previous Campbell diagram ....................................... 62
Table 4.8 – Assembly between global and local dofs of the Beam mesh ............................................. 65
Table 4.9 – Numeric data of the beam .................................................................................................. 65
Table 4.10 – Transient analysis and force impulse data ....................................................................... 66
Table 4.11 – Assembly between global and local dofs of the Monorotor mesh .................................... 69
Table 4.12 – Numeric data of the Monorotor ......................................................................................... 70
Table 4.13 – Relevant data for the frequency domain force identification method ............................... 70
Table 4.14 – Relevant data for the time domain force identification methods ...................................... 70
Table 4.15 – List of the known and unknown nodes ............................................................................. 70
Table 4.16 – Numeric data of the Propfan ............................................................................................ 76
Table 4.17 – Error computed relative to the values obtained with the author’s finite element model ... 77
Table 4.18 – Transient analysis and mass unbalance data .................................................................. 80
Table A.1 – source: (7) - edited ............................................................................................................. 94
Table A.2 – Nodal displacements .......................................................................................................... 96
Table A.3 ............................................................................................................................................... 98
viii
NOMENCLATURE
Greek Letters
Modal damping
Generic constant
Virtual work
Error vector
Euler angles
Modal eigenvalue
Poisson ratio
Constant pi
Density
Modal vector
Angular frequency
Latin Letters
Cross-sectional area of the beam element
Residue matrix
ix
Gyroscopic term
Acceleration vector
Shear effect correction factor
Distance between the mass unbalance position and the center of the shaft
Young modulus
Force vector
Generic constant
Imaginary unit
x
Stiffness of the system
Sample number of the discrete state-space system
th th
Stiffness matrix component at the i line and j column
Length
Partial length of the shaft
– lead of the ISF
Mass matrix of the system
Mass unbalance
Shape function
Number of outputs
Amplitude of the vibrations along the generalized coordinate
Principal coordinate displacement amplitude
Principal coordinate displacement vector
DMISF Delay
th
i generalized coordinate
̇ th
i generalized coordinate derivative
Radius
Time
xi
th
Time at the k sample
Input vector
State vector
Response Vector
Output vector
Abbreviations
BW Backward whirl
Dof(s) Degree(s)-of-freedom
FE Finite elements
FW Forward whirl
RB Rigid Body
RR Rayleigh-Ritz
Synch Synchronous
xii
1 INTRODUCTION
1
Since the Jeffcott publication was the first most spread work on this subject, a rotor model
consisting of one disk on a constant circular cross-section shaft, and with two rigid bearings at each
shaft end is called a Jeffcott rotor, although (as it will be seen later) in this work one has opted to
include such rotors under the designation Monorotor.
Soon, the scientists and engineers who have dedicated their work to the study of the dynamics
of this kind of structures, noted that the dynamics of a rotor is rather complex when we compare it for
instance to a spring-mass system. Since then Rotordynamics has been densely studied and
nowadays there are much broader applications in various rotor models of higher complexity than the
Föppl-Jeffcott rotor, and there is a more comprehensive knowledge of the dynamics involved.
The complexity of this dynamics can be confusing for those who venture for the first time in the
subject. Fortunately Dr. Frederick Nelson has turned it much easier with his “Rotor Dynamics without
Equations” [1]. This work by putting the focus more on the physics of the problem than in math,
becomes in my opinion an excellent introduction on the subject, and was important to clarify various
complex concepts, that were introduced in the course of this work.
One of the researchers who have devoted their work to the understanding of Rotordynamics
was D. J. Ewins. He has published recently an article where a very good review of the dynamic
problems associated with the turbomachinery of aircraft engines is included [2]. This article doesn’t
limit itself to the study of instability phenomena caused by the intrinsic structural nature of the rotor,
rather it includes a review on instability phenomena caused by the structure-aerodynamic interaction.
Another application example of Rotordynamics is presented in the work of J. J. Sinou et al. [3].
It includes a numeric prediction and an experimental validation of the non-linear dynamic behavior of a
Turbofan engine.
To gain physical intuition of this subject Nelson’s article is a good start, but to have a more
consistent work model to take proper engineering decisions, numeric methods are necessary. Maurice
Lalanne and Guy Ferraris have made this with their “Rotordynamics Prediction in Engineering” [4].
Their models and their numerical approaches were broadly used in the chapters of this work
dedicated to Rotordynamics analysis. Their work starts by defining the most common rotor models
with the help of the Rayleigh-Ritz method. Methods for the computation of the natural frequencies and
of the critical operation points are presented for each example. This work includes additionally numeric
predictions of the rotor response due to the action of various load configurations present in typical
rotor operation. Moreover the orbits described by the rotor in its vibration modes are also depicted for
each example, including as well a study about how these orbits change with changing bearing
properties. Everything is complemented with the application of the finite element method to the rotor
examples.
The book offers ultimately an application of these concepts to real case-studies where rotors
are used. In particular a Propfan engine developed in the scope of an European research project was
analyzed on the light of the concepts that were presented in the initial chapters of the work.
The studied Propfan engine will be reused on this work, not only to reproduce the associated
finite element model and resulting diagrams, but also to apply some new concepts of force
reconstruction in the diagnose of a common situation in jet engines: mass unbalance on the disks.
2
To achieve such a degree of understanding about the Propfan engine, a proper revision on
the matter was made. A short history about the Propfan research project can be found in [5].
Additional data was also found in the paper by Lalanne and Ferraris [6], where the followed models,
as well as the resultant solutions for the Propfan engine are condensed.
Also in Guy and Lalanne’s book, methods for unstable operation prediction are presented,
namely the Routh-Hurwitz method [4].
A more comprehensive and well documented work by Maurice Adams [7], “Rotating
Machinery Vibration”, has also been a source of information for this thesis. It was valuable mainly
because it offers additional and more complete information about rotor self-excitation behavior, than
the work of Lalanne and Ferraris. On the topic of dynamic analysis, this book has detailed many
specific parameters that have influence in the dynamics of rotors, so that more reality-accurate rotor
models could be programmed. They are also much more incisive in the modeling of the various
bearing characteristics, such as shaft/bearing misalignment, bearing type, and influence of lubrication.
It even contains correction factors used to predict fluid-structural interaction in turbo machines. It is
also much more detailed in several particularities of the numerical implementation of Rotordynamics,
namely in what concerns result tolerances and typical numeric errors. The book has also a dense
review on the data acquisition and usual experimental procedures in the context of rotor vibration
analysis and monitoring. The book ends with a series of interesting case studies where all the above
mentioned aspects are applied into real rotating machinery.
In what concerns Rotordynamics publication in Portuguese language, an academic book, with
many examples of application of Lalanne and Ferraris’ models was found [8]. This book was of great
importance due to the relevancy and quantity of its examples to understand the influence of the
bearing properties on the whirl motion and sense, and on the instability thresholds determination of the
rotors.
Despite they weren’t use on this work, two publications about the determination of the bearing
properties from experimental measures [9] and finite element method solutions [10] were referred.
Their connection to Rotordynamics and force reconstruction justifies the relevancy of this inclusion.
3
Efforts in the last few decades are being made in the sense of achieving a practical method
capable of solving in a satisfactory way the inverse problem of dynamic systems [11]. The concept of
inverse problem is very simple, it is just based on the inversion of the conventional determination of
response from given loadings method (dubbed the forward or direct problem). The problem is, that
response is not always easy to characterize nor the accuracy of the system data is sufficient to obtain
a satisfactory result. In fact the inverse problem is more sensitive to inaccuracies in the model’s data
than the direct problem [12].
The Frequency Domain method (FD) is, at the present time, one of the most known
techniques in the force identification field. Hundhausen’s work contains a very good review in the
subject [13]. It makes use of the measured response in several points of the system and of the
frequency response function to compute an estimation of the applied forces.
A recent submitted article developed by the IDMEC Mechanic Engineering department in
Instituto Superior Técnico, extends this method to the localization of loads in a structure [14], [15]
making use of the transmissibility concept, based on the works of Maia et al [16] [17]. This will be used
in this work in a practical example applied to Rotordynamics.
Despite the fact of obtaining good results in many cases, the FD method is not suitable for
real-time force estimation on structural systems. In fact, when there is only relatively short duration
data available the FD method renders inaccurate results. Furthermore in several applications a
representation of the force evolution over time is pretended. The force identification in time domain
has been less studied as its frequency domain equivalent, but that is beginning to change, and there is
already a broad number of published works on the subject.
The inverse problem is posed differently in the time than in the frequency domain case. This
means that the faced challenges come from other sources previously unknown that require solution,
but it can also lead to more robust and accurate force identification if these difficulties are overcame.
The sum of the weighted accelerations technique (SWAT) is one of the time domain methods
present in the literature. It was first presented by Carne et al. [18], despite the fact it was first
developed previously by Priddy and Smallwood [19]. This method makes use of modal filtering to
estimate the applied forces on the structure. Firstly a set of multiple measures of the structure
response is obtained and then the resultant rigid body modal accelerations (the accelerations on the
center of gravity of the structure) are isolated through the application of a weighting matrix. If the
inertial properties of the structure are known the resultant forces or moments corresponding to the
translations or rotations defined by the isolated rigid body mode of the structure are determined.
This method pose several difficulties for force identification, although reconstruction is
possible, one can only determine the resultant inputs in the center of gravity of the structure, so in
many applications is impossible to deduce the application point of the forces. Additionally, there are
limitations related with the modal filtering process. This procedure requires that the used sensors are
well placed and in sufficient number in order to separate the response due to rigid body modes of that
related to flexible modes of the considered structure. Since a structure has an infinite number of
modes, one will always obtain an approximate solution, being the frequency band at which the rigid
body modes can be isolated wider, as the sensor set increases.
4
An extension to the SWAT method that allows the simultaneous determination of multiple
forces acting on a structure has been developed by Genaro and Rade [20]. This was not whatsoever
included in this work.
Other technique present in the literature is based on the inversion of the equations of motion
of the system. Kammer and Steltzner [21] have presented a method that uses this principle, named
the inverse structural filter (ISF). In this method the discrete time equations of motion are inverted, this
approach is more advantageous than the one that involves the inversion of the continuous time
equations of motion, since it avoids the integration and differentiation of the measured responses.
However, the signal processing should be the aim of extra careful handling. The authors begin by
representing the direct problem in the discrete state-space form, and propose a possible way of
representing the inverse system also in the discrete state-space form. The original formulation of the
ISF uses the measured responses to reconstruct the force input in the same time instant. This
procedure has therefore a causal relationship. In both the discrete and the continuous cases, the
inverse system is contaminated with unstable poles that yield to erroneous solutions of the estimated
forces. So, besides this algorithm, Steltzner and Kammer [21] presented an approach that creates a
stable ISF directly from the measured responses to reconstruct the force input. They also introduce
the concept of ‘non causal lead’, this means that the force input is reconstructed using responses from
future time instants. This approach has shown good results in the sense that better ISF stability is
achieved, but the force reconstruction become now only close to real-time determinable.
Recently Allen and Carne [12] presented an extension to the ISF that makes use of the non-
causal concept. The delayed multi-step inverse structural filter (DMISF) pretends to improve the state-
space representation of the original formulation of the ISF, using multiple time future response
samples to reconstruct the current force input and has shown a good performance in augmenting the
stability of the solution. In this work Allen and Carne used a state-space model built only from modal
parameters, which makes force reconstruction application to experimental cases much easier. Other
method to determine the state-space matrices from the typical mass, damping and stiffness matrices
is presented in the work of Unger and De Roeck [22].
At the time this was written no direct application of the ISF method to rotating systems was
found. The works found were either applied to simple spring-mass systems, beams, or other more
complex situations, such as the simulation of the docking process of the Space Shuttle in the MIR
space station [21].
Despite of this, there are already some publications on the subject of Rotordynamics force
identification. Zutavern and Childs have developed a force identification algorithm using magnetic
bearings [23], in its turn Verhoeven has published a work in this field making use of analytical
synthetized transfer functions [24]. Spirig and Staubli [25] have presented a method that determines
the forces acting in the rotor’s seals integrating the circumferential pressure distribution measured in
the seal coordinate.
In this work it is only performed the application of the FD, SWAT, ISF and DMISF methods to
Rotordynamics examples.
5
2 FUNDAMENTALS
6
The subject of self-excited vibrations will be treated in Appendix 2 to this work, since it is a
secondary aspect to this Thesis. Self-excited vibration is an unstable state of a structural system and
is responsible for severe machine damage due to the high vibratory levels that characterize this state.
Figure 2.1 shows the evolution of the amplitude of vibration of an unstable structural system
subjected to an initial disturbance.
Figure 2.1 – Amplitude evolution with time on an unstable structural system subject to an initial disturbance –
source: [7]
The ever growing amplitude of an unstable system raises the vibratory levels to intolerable
values. In the context of rotor lateral vibration, Crandall in 1983 [7], has shown that instability is
caused by dynamic forces which are perpendicular to the instantaneous radial displacement vector of
the rotor’s orbit. These aspects will be fully characterized and described on Appendix 2, and the non-
conservative nature of these destabilizing forces will also be referred. Finally, also in this appendix, a
method of predicting the stable boundaries in terms of the rotor rotating speed will be deduced.
Let us now proceed with the presentation of each of the elementary parts that constitute
rotors.
( ̇
) (2.1.1.1)
In which denotes the number of the degree-of-freedom, the system’s generalized coordinates and
the generalized loads acting on it.
Thus the disk, shaft and mass unbalance kinetic energy must be computed, as well as the
strain energy of the shaft since it is the only rotor component considered as flexible.
In the subsequent pages is presented the detailed depiction of each rotor component and the
computation of the energies associated with each of them.
7
1. The Disk
The disk is considered rigid. This implies, since no strain energy is definable, that kinetic
energy is the only energy characterizing this component.
Figure 2.2 – The disk and its reference frames – source: [4]
Considering the disk represented in Figure 2.2, the angular speed vector [ ]
defined about its center of mass in the reference frame fixed to the disk, is as follows:
̇ ̇
[ ] [ ̇ ̇ ] (2.1.1.2)
̇ ̇
where and are the Euler angles and ̇ ̇ ̇ are its first time derivatives, dubbed rate of nutation,
rate of spin and rate of precession respectively.
Moreover, designating the displacements of the disk’s center of mass along the directions
and of the inertial reference frame respectively by and , the kinetic energy expression for the disk
results:
( ̇ ̇ ) ( ) (2.1.1.3)
is the disk’s mass, and , , , represent the disk’s inertia about its principal axis .
The expression (2.1.1.3) can be simplified, noting that . Further simplification can be
achieved considering the angles and very small and constant rate of spin ̇ in (2.1.1.2).
Substituting these statements in expression (2.1.1.2) and then in (2.1.1.3) it results:
̇ ̇ ̇ ̇
[ ] [ ̇ ̇ ] [ ̇ ]
̇ ̇ ̇ ̇
8
( ̇ ̇ ) ( ̇ ⏟ ̇ ̇ ̇ ̇ (⏟̇ ) (⏟ ̇ ) ̇ )
( ̇ (⏟ ̇ ) )
Giving the following expression for the kinetic energy of the disk:
( ̇ ̇ ) ( ̇ ̇ ) ( ̇ )
(2.1.1.4)
where, the last term ̇ , is accounting for the Gyroscopic effect.
2. The Shaft
The shaft will be considered as a flexible beam with circular uniform cross-section, thus it is
characterized both by kinetic and strain energies. In the Figure 2.3, the shaft cross section is shown
along with two reference frames, the inertial ones with displacements and along the and axis,
and the rotating reference frame solidary with the shaft’s cross-section center, characterized by the
displacements and along the and axis.
The kinetic energy for such a shaft can be described exactly by the same course of reasoning
followed for the disk, with the slightly difference this time, that it should be applied to an element of
length of constant cross-section. The kinetic energy expression for the shaft is:
∫ ( ̇ ̇ ) ∫ ( ̇ ̇ ) ∫ ̇
(2.1.1.5)
Here denotes mass per unit volume, is the shaft’s cross-sectional area which is supposed
to be constant, and is the area moment of inertial of the shaft cross-section about its neutral axis,
9
which is once more considered as constant. As to the physical meaning of the terms of the equation
(2.1.1.5), it can be said that the second one is the secondary effect of rotary inertia, coherent with
Timoshenko beam theory, and the last term represents the gyroscopic effect.
Considering the strain caused by the rotation of the shaft, and the strain caused by a constant
axial force applied in the shaft’s ends, and if additionally the shaft is considered elastic, i.e. it is
inside the scope of validity of the Hook’s Law, it results for the strain energy of the symmetric shaft:
[4]
∫ [( ) ( ) ] ∫ [( ) ( ) ]
(2.1.1.6)
The first term in the expression (2.1.1.6) accounts for the strain energy resultant from the
longitudinal deformation of the spinning shaft. This longitudinal strain due to rotation of the points
inside the constant cross-section of the shaft is considered approximately linear, since the
displacements are small and the cross-section has symmetry relative to the rotating reference frame.
The longitudinal strain caused by the presence of an axial force has only non-linear terms and the
associated strain energy is included in the second term of the equation (2.1.1.6). Constant, is the
Young’s Modulus of the shaft’s material.
The bearings will be modeled as represented in the Figure 2.4, the next step is to write the
virtual work of the forces acting in the springs and dampers constituting them, in order to relate the
components and of the force with the bearing’s characteristics.
10
The virtual work done by the forces acting on the bearings is given by:
(2.1.1.7)
The virtual work done by the forces acting on the springs and dampers of the bearing can be
written as:
̇
̇ ̇ ̇ (2.1.1.8)
Joining (2.1.1.7) and (2.1.1.8) together and writing the resultant expression in matricial form,
give us after elimination of the virtual displacements:
̇
[ ] [ ][ ] [ ][ ] (2.1.1.9)
̇
4. The mass unbalance
The mass unbalance will be modeled as a punctual mass situated at a given distance
from the center of the shaft.
With the help of Figure 2.5 let us define first the displacement vector , in the inertial
reference frame :
| | (2.1.1.10)
̇
| | (2.1.1.11)
̇
11
With (2.1.1.11), and observing that the punctual mass is much smaller than the disk’s
mass and thus can be neglected, the kinetic energy of the punctual mass yields:
( ̇ ̇ ) (2.1.1.12)
( ) (2.1.2.1)
12
Figure 2.7 – Inertial reference frame coordinates – source: [4]
( ) ( ) (2.1.2.2)
( ) ( ) (2.1.2.3)
Where and denote the displacements in the and directions on the first rotor, and
are generalized independent coordinates, and ( ) denotes the shape equation for the first rotor,
and coincides with the definition in the equation (2.1.2.1).
If the angular displacements and of the first rotor, shown in Figure 2.7 are considered
small, the following relations are valid:
( )
( ) ( ) (2.1.2.4)
( )
( ) ( ) (2.1.2.5)
and: ( )
The needed second derivate of and to compute the strain energy, are given by:
( )
( ) ( ) (2.1.2.6)
( )
( ) ( ) (2.1.2.7)
and: ( ) ( )
The second rotor, the outer one, is assumed to be rigid, so the second rotor displacements (
and ) and angular displacements ( and ) are written in the following way:
( ) ( ) ( ) (2.1.2.8)
13
( ) ( ) ( ) (2.1.2.9)
( )
( ) ( ) (2.1.2.10)
( )
( ) ( ) (2.1.2.11)
It is observable from (2.1.2.9) and (2.1.2.10) that is a constant, since the second shaft is
rigid. For this reason there is no definable strain energy for the second rotor, in conformity with the
rigid rotor assumption:
(2.1.2.12)
Applying expressions (2.1.2.1-5) and (2.1.2.6-11) in the kinetic energy equations (2.1.1.4) and
(2.1.1.5), for each of the two shafts and two rotors, it results:
( ̇ ̇ ) ̇ ̇ (2.1.2.13)
with:
( ) ( ) ( )
∫ ( ) ∫ ( )
∫ ( ) ∫ ( )
(2.1.2.14)
( ) ∫ ( ) (2.1.2.15)
∫ ( ) (2.1.2.16)
where is dubbed gyroscopic term, the index 1 and 2 concern rotor 1 and 2 respectively, and where
the constant terms of equations (2.1.1.4) and (2.1.1.5) were ignored since they will disappear anyway
after application of Lagrange equation.
The rotating speeds of the rotors are related by a constant ratio , which is integer, following
the relation:
(2.1.2.17)
14
In what concerns on this case the strain energy, the only contribution comes from the first
rotor. Recalling equation (2.1.1.6), substituting on it the expressions (2.1.2.5) and (2.1.2.6), and
considering the axial force null, it yields:
( ) (2.1.2.18)
where:
∫ ( ) (2.1.2.19)
Substituting expressions (2.1.2.1) and (2.1.2.2) in equation (2.1.1.9), and noting from the
Figure 2.6 that the bearing is located at a distance from the reference frame origin, the components
of the generalized force acting in the bearing become:
̇
[ ] ( )[ ][ ] ( )[ ][ ] (2.1.2.20)
⏟ ⏟ ̇
[ ] [ ]
Finally, the kinetic energy due to the presence of a punctual mass in a disk and at a given
distance from the axis of rotation of the Multirotor is written with the help of equation (2.1.1.12). If the
punctual mass is applied to the first disk in , its kinetic energy is given by:
( )( ̇ ̇ ) (2.1.2.21)
( )( ̇ ̇ ) (2.1.2.22)
If and are two generic transversal forces applied in a specific longitudinal coordinate of
the rotor, and acting along the and directions respectively, the virtual work done by these forces is
given by:
( ) ( ) (2.1.2.23)
Which after can be rewritten in the generalized coordinates in the following way;
( ) ( ) ( ) ( ) (2.1.2.24)
( ) ( )
( ) ( ) (2.1.2.25)
15
So, for the most general case of two coaxial rotors with a viscously damped bearing, rigid disk,
symmetric shaft, with external transversal applied forces ( ) and ( ), Coriolis acceleration effect
and assuming no motion in the axial direction, the resulting motion equations, considering only the
degrees-of-freedom (Dofs) perpendicular to the axial axis are given by:
̈ ( ) ̇
[ ]{ } ([ ] [ ]) { } [ ]{ }
⏟ ̈ ⏟ ( ) ̇ ⏟
[ ] ⏟ ( ) [ ]
[ ]
{ ( )}
(2.1.2.26)
In the equation above, is the inertial matrix, and the gyroscopic effect on rotor 1 and 2
respectively, is the rotating speed, the inner-outer spool rotating speed ratio, [ ] include the
damping coefficients in the bearings, [ ] the total stiffness matrix (bearings plus strain stiffness), and
are the applied forces along the two defined generalized coordinates and .
Despite the previous deduction has been applied to a Multirotor configuration, it is easily
adapted to a Monorotor configuration if the outer rotor terms in (2.1.2.26) are considered null.
(2.1.3.1)
(2.1.3.2)
with:
√
Where denotes the angular frequency of the free vibrations and denotes the modal
damping associated with each free vibration mode. Of course is null in undamped rotors. is the
imaginary unit defined as √ . denote the amplitude of the vibrations in the generalized
coordinates .
Introducing these equations in (2.1.2.26) it results the following eigenvalue/eigenvector modal
problem:
16
[ ]{ } { } (2.1.3.3)
( ) ( ( ) )
( ( ) ) ( )
(2.1.3.4)
The solution of the equation (2.1.3.4) leads to two pairs of complex conjugate roots,
and , where and are the first and the second natural
√ √
frequencies of the rotor and are dependent of the rotating speed of the inner shaft ( ), and and
are the associated modal damping. This procedure is valid whether or not viscous damping is present.
These results yield to the equations that are used to obtain the Campbell Diagram. The Campbell
Diagram, exemplified on the Figure 2.8, is a graphic that show the variation of the natural
frequencies of the rotor with increasing rotating speed in turn of its axis, i.e. the frequency curves.
Alongside this representation, several lines can be traced from the graphic’s origin with an inclination
that establishes a relevant relation between the rotating speed and the natural frequencies. These
lines are generically defined by , where is the inner-outer spool rotating speed ratio and
the rotating speed in rotations per minute. The most common line traced on the Campbell Diagram
is the synchronous line, i.e. the line that includes the points at which the natural frequency equals the
rotating speed, and corresponds to the line defined by on Figure 2.8 The knowledge of the
intersection points of these lines with the frequency curves is very relevant for the prediction of critical
operation points as it will be shown later.
17
2. Response to forces due to mass unbalances
Similarly to the description in the point 4 of the section 2.1.1, the effect of a punctual mass
‘ ’ at a given distance ‘ ’ from the center of the disk ‘ ’ in the force vector is:
( ) ( ) ( )
{ } { } (2.1.3.5)
( ) ( ) ( )
in which ( ) is the value of the displacement function in the location of the disk.
Considering additionally that viscous damping effects are neglected ([ ] [ ]) the solutions of
the resulting equation are sought in the following form:
(2.1.3.6)
(2.1.3.7)
Substituting (2.1.3.4), (2.1.3.5) and (2.1.3.6) in (2.1.2.26), and considering that the unbalance
mass is situated in the disk 1 situated at , it results:
( ) ( )
[ [ ] ( [ ]) [ ]] { } { }
( ) ( )
(2.1.3.8)
Solving equation (2.1.3.8) for the variables and , it yields:
( ) ( ( ) )
(2.1.3.9)
( )( ) ( )
( ) ( ( ) )
(2.1.3.10)
( )( ) ( )
If the unbalance mass is situated in the disk 2 at , the speed in the solution’s equations
(2.1.3.6-7) must be the rotating speed of the respective shaft. Since disk 2 is located in the outer shaft
is the proper speed to use. The modal vector equations are then obtained by following the
procedure used before, and are shown just below:
( ) ( ( ) )
(2.1.3.11)
( )( ) ( )
( ) ( ( ) )
(2.1.3.12)
( )( ) ( )
If (asymmetric rotor), then one can conclude that the rotors describe elliptical
orbits; otherwise the orbits are circular. If the denominators are made equal to zero, we can obtain the
critical speeds by solving the resultant equation. There are two critical speeds when the rotor is
asymmetric and only one solution when the rotor is symmetric.
18
Since an unbalance on the inner disk is synchronous with the rotation of the inner shaft, these
critical speeds will correspond to the intersection points between the synchronous line and the
frequency curves on the Campbell Diagram if disk one is unbalanced. But if the unbalanced mass is
applied on disk 2, the imparted force can be asynchronous in relation to the inner shaft depending on
the relation between rotating speeds quantified by . So the critical speeds of such an unbalance will
correspond to the intersection points between the dependent asynchronous line and the frequency
curves on the Campbell Diagram.
If viscous damping is added to the model then the solutions can be sought in the following
form:
(2.1.3.13)
(2.1.3.14)
In its turn, using (2.1.3.13), (2.1.3.14) and (2.1.3.5) in the equation (2.1.2.26), the values of the
coefficients , , and are obtained by solving the resultant system.
The orbits will be elliptical in case of asymmetric stiffness in the bearings but the existence of
damping animates this ellipse with a precession motion. This means that the semi-major and semi-
minor axis of the ellipse will make an angle relative to the inertial reference frame axis.
Caution must be taken in the used value of the speed in the solution equations (2.1.3.13-14), if
the unbalance mass is placed on disk 2 then again is the rotating speed to use.
The Campbell Diagram can be used on the prediction of the critical speeds for this case
exactly in the same way as it was described previously, attending, of course, if the considered
excitation force is synchronous or asynchronous, i.e. affected by the factor .
Equations (2.1.3.9-14) can now be used to obtain the response diagram of the rotor with the
respective bearing conditions. The response diagram in Multirotor configuration is a plot of the
logarithm of the sum of the transversal vibration amplitude( ) against the rotating speed of the
inmost shaft, as exemplifies Figure 2.9.
19
2.1.4 Rotordynamics Modal Orbits and Sense of the Whirl
In the last section rotor vibration orbits were characterized for each Multirotor type.
Nevertheless, nothing was said about the sense of the orbit described by the rotor while it vibrates and
rotates simultaneously. In fact, as it will be addressed shortly, rotor orbits are not trivial phenomena. In
some rotors the transversal orbit described by the shaft vibration has an opposite sense to that of the
rotor’s rotating sense, it is then said that the rotor is in backward whirl. If the shaft vibration orbit has a
corotational sense with that of the rotor’s rotating speed, then the rotor motion is dubbed as forward
whirl.
On this section, it will be analytically shown, the existence of this phenomena as well as what
causes the rotor to be in backward whirl instead of forward whirl and vice-versa.
Let us first consider again the Multirotor represented in Figure 2.6. As previously discussed,
the displacements of the rotor in the and directions are given by and expressions such as
that displayed in the previous section. It was already mentioned in that section that the orbits could be
circular or elliptical depending if the rotor was symmetric or asymmetric.
Figure 2.10 (a-c) shows the various possible orbit shapes in Rotordynamics, summarizing also
the bearing characteristics behind each one of them.
(a) (b)
(c)
Shaft position
Now let us show the way the sense of whirl is determined. For this let us consider the
displacement vector ⃗⃗⃗⃗⃗⃗ of the rotor rotating axis center relative to the inertial reference frame origin,
displayed in the Figure 2.11.
20
A Shaft position at 𝑦 𝑙
⃗⃗⃗⃗⃗⃗ [ ] (2.1.4.1)
Consequently the displacement vector derivative yields, making use of the chain rule:
⃗⃗⃗⃗⃗⃗ [ ] (2.1.4.2)
The sense of whirl is deduced from the signal of the crossed product between the
displacement vector and its derivative:
If the resultant vector in (2.1.4.3) is pointing to the positive semi axis of , then the rotor is
said to be in forward whirl (FW). On the contrary if it is pointing the other way around it is said to be in
backward whirl (BW).
To help the visualization of this phenomenon Figure 2.12 shows position of an unbalanced
mass relative to the inertial axis as the disk of the Monorotor there represented follows the path both
in forward and backward whirl. This depiction is only valid if the disk’s rotating speed is considered to
be equal to the orbit speed, i.e. synchronous. This assumption is consistent with what was predicted
on the last section. As we saw there, for a mass unbalance on the inner rotor’s disk of a Multirotor
configuration, whose vibration modes are given by the solutions (2.1.3.6-7), the orbit revolution speed
is synchronous with the disk’s rotating speed .
21
Figure 2.12 – Position of a mass unbalance (in bold) in a disk performing Forward and Backward Whirl
The expressions of and will of course depend of the bearing characteristics of the rotor.
Anyway, only for exemplificative purposes the determination of the whirl sense for a particular case of
the undamped Multirotor like the one shown in Figure 2.6 with asymmetric stiffness bearing properties,
and a mass unbalance on disk 1, is presented in the Appendix 1 to this work.
[ ]{ ̈ } [ ]{ ̇ } [ ]{ } { ( )} (2.2.1.1)
22
Where [ ] is the mass matrix of the system, [ ] is the damping matrix, [ ] the
stiffness matrix of the structure and the vector { } { } is the generalized coordinate
vector, being a positive integer representing the number of degrees-of-freedom (dofs) of the finite
element model. Vector { ( )} includes the loads exerted in the system’s dofs.
Equation (2.1.2.26) for instance, is a particular application of equation (2.2.1.1) to Multirotors.
The Mono- and Multirotor dynamic system’s matrices , and are computable through application
of the Rayleigh-Ritz method as shown in chapter 2.1.2, or by using the finite element model as it will
be shown later on the chapter 3.1.
For each frequency value, the response in the dof in steady-state conditions can be written
as:
(2.2.1.2)
Applying the first and second derivatives with respect to time, one obtains respectively:
̇ (2.2.1.3)
̈ (2.2.1.4)
[
⏟ [ ] [ ] [ ] ]{ } { ( )} (2.2.1.5)
[ ( )]
where: { } { }
Observing equation (2.2.1.5) one should notice that the relation between the loads acting on
the system and the caused displacements is only dependent of the matrix [ ( )] . Due to its
determinant importance the [ ( )] matrix is known as the dynamic stiffness matrix.
If the dynamic stiffness matrix is computed the force reconstruction becomes possible in every
node simply by multiplying it by the dofs measured response as the following equation shows.
{ ( )} [ ( )]{ } (2.2.1.6)
Despite its simple aspect, equation (2.2.1.6) is hard to apply to complex systems. In the
majority of the applications there are a large number of dofs, or some measurement points are simply
inaccessible making the measurement of all the system’s dofs unpractical. Therefore a way of
determining the unknown system’s dofs response using the known ones is vital. The Transmissibility
concept in multi-dof systems is the tool that supplies that result. This means that measurements of the
response done in some few nodes are used to deduce the response in the dofs where the response is
unknown. A recent approach made by [15] allows not only the deduction of the responses in all the
23
structures dofs, but also the localization of the loads which are acting in the structure. This approach
will be briefly introduced in the end of this section.
Even knowing all the system’s dofs responses, it is not practical to reconstruct all the system’s
load vector. Instead it is more efficient to reconstruct the loads in the spots where we are sure they are
acting. So if the location of the applied loads is known, the problem can then be reduced to the
following equation:
{ ( )} [ ( )] { } (2.2.1.7)
where is the number of force inputs and the number of measured dofs. [ ( )] in this case is
a reduced dynamic stiffness matrix that relates only the involved dofs.
Equation (2.2.1.7) is solvable as long as the number of used responses is larger or equal to
the number of Forces acting on the structural system.
{ ( )} [ ( )]{ ( )} (2.2.1.8)
where: [ ] [ ] is the receptance matrix defined as the inverse of the dynamics stiffness
matrix.
Therefore, considering equation (2.2.1.8), { }, { } are related with { } by:
{ } [ ]{ }
{ } [ ]{ } (2.2.1.9)
Where the indices and refer to the established relation between the response
coordinates and , and the load’s coordinates respectively.
Solving equations (2.2.1.9) by relating each one of them through the load vector { }, it
results:
24
{ } [⏟ ][ ] { } (2.2.1.10)
[ ]
in which [ ] designate the pseudo-inverse matrix and where [ ] is the transmissibility matrix. This
matrix establishes a relation between the dofs responses under a specific load condition . The
proposed force localization method works by assuming random and different combinations for the
force point of application and to compute the resulting transmissibility matrix for each combination
since, as it was already mentioned, the matrix matrices are assumed to be known. After equation
(2.2.1.10) is applied in a user defined frequency range and a result is obtained for the response
vector { ̃ }. The symbol ̃ was added, to distinguish between the computed response vector from the
actually measured response vector { }.
For each combination of forces the accumulated error between vectors { ̃ } and { } is
computed component by component as the next expression states, and saved in an array :
∑ ( ( ( ̃ ( ))) ( ( ( )))) { } { }
(2.2.1.11)
The combination at which the error is minimized is expected to correspond to the real point of
application of the forces.
After this is accomplished the amplitude of the loads can be determined through equation
(2.2.1.7), keeping on mind that this is possible, as long as the number of used responses is larger
or equal to the number of forces acting on the structural system.
Transformation of the Load Vector from the Frequency Domain Method to the Time-Domain
If is intended a comparison between the force reconstruction obtained using the FD with the
results obtained by the time-domain methods presented below, an inverse FFT (Fast Fourier
Transform) can be applied to load vector estimated by the FD method in order to convert frequency-
domain results into time-domain results.
25
Figure 2.13 illustrates how a spectrum of a response signal can be obtained from the FRF
(Frequency Response Function) through the discrete sum of individual sinusoidal modal contributions.
This is the procedure followed by the inverse FFT. Likewise, if it is applied to the load vector in the
frequency domain obtained in (2.2.1.7), a time-domain representation of the loads can be achieved.
{ } [ ] { } (2.2.2.1)
where the sought weighting matrix [ ] is used to extract the vector { } of rigid
body accelerations, being the number of rigid body modes, and the number of measurement
points.
The method is completed after multiplying the rigid body accelerations by the rigid body mass
properties to obtain the load vector { ( )} .
{ } { ( )} (2.2.2.2)
If this method is systematically applied to the measurements at each time instant, a time –
domain force reconstruction at the structure’s center of gravity is obtained.
Let us now introduce how the weighting matrix [ ] is computed.
Firstly, it begins by establishing a relationship between the measured accelerations and the
modal shapes using a sum of modal contributions representation.
26
{ } [ ]{ ̈ } (2.2.2.3)
where { } is a vector that contains the measured accelerations at the measurement points, { ̈ }
is the vector of modal displacements accelerations, [ ] is a matrix containing the mode
shapes, is the number of modes (both rigid body and flexible).It should be remarked that, according
to vibrations theory, any continuous system has an infinite number of modes, meaning that (2.2.2.3) is
just an approximation. Matrix [ ] can either be computed solving the Eigen problem if a finite element
model of the structure is available or through the algorithm of mode isolation (AMI) [27] if an
experimental mount is being analyzed.
Equation (2.2.2.3) can be rewritten in a more convenient way, if the rigid body mode vectors
are considered mass normalized, i.e. the modal vectors { } are such that make { } [ ]{ } [ ] and
{ } [ ]{ } [ ] , being [ ] the identity matrix and the eigenvalue, and if they are
assigned to the leading columns of [ ], it results:
{ }
{ } [[ ] [ ]] { } (2.2.2.4)
{ ̈ }
where the and indices correspond to the rigid body and elastic modes, respectively and { ̈ }
represent the second derivative of the elastic modal displacements which will multiply by the [ ]
matrix columns. If (2.2.2.1) is introduced in (2.2.2.4), and taking into account once again that the
weighting matrix must isolate the rigid body modes, it yields:
[ ] [[ ] [ ]] [[ ] [ ] ]
[ ]
[ ] ([[ ] [ ]] ) [ ] (2.2.2.5)
[ ]
It should be noted, that as long as the condition is verified, the columns of the
weighting matrix are able to extract the rigid body accelerations from the modal data.
After [ ] is determined equations (2.2.2.1-2) are ready to be computed, giving the load time-
history of the resultant in the structure’s center of gravity of the loads corresponding to the rigid body
motions introduced in (2.2.2.5) in the [ ] matrix.
To conclude this section it is worth mentioning that a SWAT variant was developed by Genaro
and Rade [20] that enhances this method allowing the possibility to determine multiple forces acting
simultaneously.
27
{ ̇ ( )} [ ]{ ( )} [ ]{ ( )}
{ ( )} [ ]{ ( )} [ ]{ ( )} (2.2.3.1)
Here, { } is the state vector of the system, { } contains the inputs acting on the system,
{ } is the vector that contains the outputs of the considered system, [ ] is the system’s
matrix, [ ] the input influence matrix, [ ] is the output influence matrix and [ ] is the
direct throughput matrix. The index denotes that the system is a continuous time state-space
system and { ̇ } is the first time derivative of vector . The first expression in (2.2.3.1) is called the
state-space equation and the second expression is dubbed the output equation.
In many applications, e.g. in experimental applications, a discrete time representation of the
state-space is preferred. The output of the discrete time system will exactly reproduce the one
obtained with the continuous time system, as long as the zero order hold (ZOH) condition is verified,
i.e. if the input is constant between samples [12]. The discrete time state-space equations are the
following ones [21]:
{ } [ ]{ } [ ]{ }
{ } [ ]{ } [ ]{ } (2.2.3.2)
The matrices and vectors of this system have an identical meaning to the ones on the
continuous time state-space system, and the index is the sample instant number. The conversion
from continuous to discrete time state-space imply a manipulation of the [ ] and [ ] matrices, while
[ ] and [ ] remain unaltered. The new matrices are displayed in the following equations. [12]
[ ] ( )
[ ] [ ] [ ( ) [ ]] [ ]
[ ] [ ]
[ ] [ ] (2.2.3.3)
in equation (2.2.3.3) is the time between successive time samples. The time at which each sample
is obtained can be thus obtained simply by multiplying by the th time step at which time is intended
to be computed.
Let us now focus on the content of the vectors and matrices of the state-space systems in the
context of the dynamic of mechanical systems.
In the most general case of a degree of freedom dynamic system, with stiffness [ ], mass
[ ], damping [ ] and with a load { ( )} acting on it, the equations of motion are:
[ ] { ( ̈ )} [ ] { ̇ ( )} [ ] { ( )} [ ] { ( )}
(2.2.3.6)
28
where { } is the modal coordinates vector of the system and where { ̈ } and { ̇ } denote the second
and first time derivates of the modal coordinate vector. [ ] is simply a matrix of ones and zeros
that selects the subset at which forces are being applied, is the number of input loads.
One possible choice for the state-space vector is the following:
{ }
{ } { } (2.2.3.7)
{ ̇}
Let us now introduce the state vector { } on equation (2.2.3.6). It returns the following
representation:
[ ] [ ] [ ] [ ] [ ]
[ ] { ̇} [ ]{ } [ ][ ]{ }
[ ] [ ] [ ] [ ] [ ]
(2.2.3.8)
If one works out this equation one obtains in the end the following expression:
[ ] [ ] [ ]
{ ̇} [ ]{ } [ ][ ]{ }
⏟ [ ][ ] [ ][ ] ⏟[ ]
[ ] [ ]
(2.2.3.9)
Comparing equation (2.2.3.9) with the state-space equation in (2.2.3.1), matrices [ ] and [ ]
become determined by similarity.
The matrices of the output equation, i.e. the second expression in (2.2.3.2), depend on the
nature of the system as well as on the intended output variable. For structural systems such the ones
studied in this work, where the force locations are not always entirely known, measures are obtained
for degrees-of-freedom on which forces are not applied, the system is therefore called a non-
collocated system.
The state-space of non-collocated systems possesses non-minimum phase zeros, this
statement implies that the direct throughput matrix [ ] is dropped rank and thus the output equation
for this systems will only depend on the output influence matrix [21].
In control theory, a system is said to be minimum-phase if the system and its inverse are
causal and stable. This is not the case of the state-space system used in force identification methods,
since they may be associated to unstable inverse systems. This essentially happens because the
zeros of the non-collocated state-space systems have positive real parts, i.e. in the right-hand side of
the complex plane. This is a condition for a system to have an unstable inverse system according to
the control theory, since the eigenvalues of the inverse system are the zeros of its associated system.
Recall from vibration theory how positive real parts in the eigenvalues are closely related to unstable
phenomena (see Appendix 2).
The output influence matrix [ ], will work simply as an output selector, in which the selection
options are the state vector components. If the displacements on the degrees-of-freedom of structure
29
are considered as the output of the system, and recalling that the nodal displacements are included in
the first positions of the state vector in (2.2.3.7), it results for the output equation:
{ } [[
⏟] [ ] ]{ } (2.2.3.10)
[ ]
This is not whatsoever the only way to determine the matrices comprising the State-space
system. For instance, Allen and Carne [26] have derived the state-space system of the structures they
have tested from modal parameters. This has been made fitting to the experimentally measured
Frequency Response Function (FRF) [27] a state-space modal model that follows the standard
definition:
[ ] [ ]
[ ( )] ∑ ( ) (2.2.3.11)
where ( ) denotes the complex conjugate, [ ( )] is the FRF matrix at frequency , is the modal
eigenvalue, i.e. √ , where is the modal damping ratio and the modal
undamped natural frequency for the th mode of vibration. [ ] is the residue matrix, defined as:
[ ] { } { } (2.2.3.12)
where { } and { } are the modal vectors of the state-space of the response and drive (load)
locations. Considering that the loads are only applied to a subset of the response locations, it can be
expressed by:
{ } { }
{ } { } [ ] (2.2.3.13)
[ ] [ ] , [ ] [ ] [ ],
[ ] [ ] (2.2.3.14)
where is a diagonal matrix containing the eigenvalues in ascending order and the columns of
contain the { } vectors in the same order as the eigenvalues. These matrices are then introduced in
the state-space equations (2.2.3.1). [ ] is zero since the system is a non-minimum phase one as was
justified before.
Now that the continuous state-space system has been completely characterized for a
structural system, and after presenting two ways of describing the system’s matrices, the topic of the
inverse state-space system dubbed the inverse structural filter will be addressed.
30
2.2.4 ISF – Inverse Structural Filter
The inverse structural filter (ISF) first deduced by Steltzner and Kammer [21], is a time domain
method which is based on the inversion of the state-space system depiction of the equations of
motion, in which the taken inputs are the structural response data and returned outputs are the
estimations of the input loads applied in a specific point of the structure. This is an advantage
compared to the SWAT method, which only computes the resultant forces and the center of gravity of
the structure.
Before inverting the system, an additional manipulation should be performed on the state-
space system. On experimental applications accelerations are more easily measured than
displacements. Therefore, the second derivative of the state and output equations will be computed in
order to obtain a state-space representation of the accelerations. They are expressed in the following
form:
{ ̈} [ ]{ ̇ } [ ]{ ̇ },
{ ̈} [ ]{ ̈ } [ ]([ ]{ ̇ } [ ]{ ̇ }), (2.2.4.1)
{ ̈} [ ]{ ̇ } [ ]{ ̇ },
with:
[ ] [ ][ ] [ ] [ ][ ]
The discrete time representation is simply obtained for the acceleration representation after
the matrix transformations in (2.2.3.3) are applied. While [ ] and [ ] remain constant and equal to
[ ] and [ ] respectively.
{ } [ ]{ ̇ } [ ]{ ̇ },
{ ̈ } [ ]{ ̇ } [ ]{ ̇ } (2.2.4.2)
Steltzner and Kammer [21] have then manipulated the discrete time state-space equations
yielding to the interchange of the input with the output, obtaining the following inverse state-space
system:
{ } [ ̂ ]{ ̇ } [ ̂ ]{ ̈ }
{ ̇ } [ ̂ ]{ ̇ } [ ̂ ]{ ̈ } (2.2.4.3)
where,
̂ ̂
̂ ̂ (2.2.4.4)
31
It should be noted that what is obtained in equation (2.2.4.3) is the time derivative of the loads,
meaning that afterwards a numeric integration method should be applied in order to obtain the load
reconstruction.
As it has been told, non-collocated state-space systems are common practice in the field of
force reconstruction. It can be shown [21] that the non-minimum phase zeros that characterize such
state-space systems, have positive real parts, and become the eigenvalues of the inverse state-space
system. This generates an unstable inverse system. Due to this fact, the formulation displayed in
equation (2.2.4.3) has not been used, and alternative ways of representation of the structure were
implemented.
Steltzner and Kammer, [21], used expression (2.2.4.5) for the pulse response of the state-
space system assuming zero initial conditions for the structural acceleration, which is presented
below:
̈ ∑ ̇ (2.2.4.5)
where the Markov parameters proposed by Steltzner and Kammer are [21]:
Equation (2.2.4.5) is derived stepping forward the expressions in (2.2.4.2), i.e. substituting
successively the first expression in (2.2.4.2), into the second expression in (2.2.4.2).
Step 1: { ̈ } [ ]([ ]{ ̇ } [ ]{ ̇ }) [ ]{ ̇ }
Step 2: { ̈ } [ ][ ]( [ ]{ ̇ } [ ]{ ̇ }) [ ][ ]{ ̇ } [ ]{ ̇ }
In the limit : { ̈ } [ ][ ] [ ]{ ̇ } [ ][ ][ ]{ ̇ } [ ][ ]{ ̇ } [ ]{ ̇ }
(2.2.4.5)
Though this expression involves a limitless number of Markov parameters, it is known [21] that
the system becomes fully described using only Markov parameters, being the number of
observable and controllable modes of the structural system.
Likewise, the same representation can be adopted for the inverse system displayed in
(2.2.4.3), again based on the assumption of zero initial conditions and following the same procedure
described for the demonstration of expression (2.2.4.5), but this time applied to the expressions in
(2.2.4.3). Following this procedure it results for the inverse system:
̇ ∑ ̂ ̈ (2.2.4.6)
The inverse system Markov parameters ̂ proposed by Steltzner and Kammer [21] are
presented below:
̂ ̂ ̂ ̂̂ ̂ (2.2.4.7)
32
This representation does not always result in a converging sum, for the same reasons
explained previously involving non-collocated systems.
In the literature, various authors as in [21] as well as in [12] have opted to neglect the direct
throughput matrix [ ] of the discrete state-space system and to step forward the time of the system
times in order to improve the ISF performance. The direct throughput matrix result from the
multiplication between [ ] and [ ], which in case modal parameters are used to the construction of
the state-space, is highly dependent on the real part of [ ][ ][ ] . This term in its turn could be very
small when applied to lightly damped structures, also meaning lack of accuracy in its determination.
Applying these changes to equations (2.2.4.2) it results in the following equations:
{ } [ ]{ ̇ } [ ]{ ̇ }
{ ̈ } [ ]{ ̇ } [⏟ ]{ ̇ }
( )
{ ̈ } [ ]([ ]{ ̇ } [ ]{ ̇ }) [⏟ ][ ] { ̇ } [⏟ ][ ] { ̇ }
[ ] [ ]
Stepping 2 time instants:
{ } [ ]{ ̇ } [ ]{ ̇ }
{ ̈ } [ ][ ]{ ̇ } [ ][ ]{ ̇ }
{ ̈ } [ ][ ]([ ]{ ̇ } [ ]{ ̇ }) [ ][ ]
⏟ { ̇ }
( )
{ ̈ } [ ][ ] { ̇ }
⏟ [ ][ ][ ] { ̇ }
⏟
[ ] [ ]
Stepping the time forward time instants it yields:
{ ̈ } [ ][ ] { ̇ }
⏟ [ ][ ]
⏟ [ ]{ ̇ } (2.2.4.8)
[ ] [ ]
The inversion of this state-space system follows exactly the same standard procedure used in
(2.2.4.3) and (2.2.4.4), with the slightly difference that instead of [ ]or [ ] the inverse system matrices
should be computed with [ ] and [ ] since the state-space system is now defined in a different way.
̇ ∑ ̂ ̈ (2.2.4.9)
with:
This equation uses the responses measured at instants to estimate the loads at a
previous instant , i.e. the inverse system associated with (2.2.4.9) is non-causal. This solution has
made results obtained with ISF stable in structures where they otherwise were divergent.
33
2.2.5 DMISF – Delayed Multi-step Structural Filter
Based on the work of Steltzner and Kammer, Allen and Carne [12] developed the Delayed
Multi-step Inverse Structural Filter (DMISF). This method improves the performance of the common
ISF since it takes into account the stabilizing effect of the non-causal ISF, i.e. an ISF that uses future
values of the response to estimate the forces at a given time. This implies that a delay is involved
between the response data acquisition and the load reconstruction.
Recalling the stepping forward procedure displayed in the expression (2.2.4.8), Allen and
Carne reformulate the output equation of the state-space system by stacking this equations for a given
number of time steps and rewriting (2.2.4.8) for a step of 1 time instant as:
{ } [ ]{ ̇ } [ ]{ ̇ }
{ ̈ } [ ]([ ]{ ̇ } [ ]{ ̇ }) (2.2.5.1)
{ ̈ } [ ][ ] { ̇ }
⏟ [ ][ ] { ̇ }
⏟
[ ] [ ]
{ } [ ]{ ̇ } [ ]{ ̇ }
{ ̈ } [ ]{ ̇ } [ ]{ ̇ } (2.2.5.2)
{ ̈ } [ ][ ]{ ̇ } [ ][ ]{ ̇ } [ ]{ ̇ }
{ ̈ } [ ][ ] { ̇ } [ ][ ] [ ]{ ̇ } [ ][ ]{ ̇ } [ ]{ ̇ })
Stacking the time output equation for each time step, it results in the following modified output
equation.
{ ̈ } [ ] [ ] [ ] [ ] { ̇ }
{ ̈ } [ ][ ] [ ][ ] [ ] [ ] { ̇ }
[ ]{ ̇ }
[ ]
{{ ̈
⏟ }} ⏟[ ][ ] [[ ][ ]
⏟ [ ] [ ][ ] [ ] [ {{ ̇
]] ⏟ }}
{ ̈ } [ ] [ ] { ̇ }
(2.2.5.3)
The state equation must be also slightly changed, in order to consider the new definition of the
input load stacking vector { ̇ }. It is called so because it literally stacks each time instant of the load
considered in the DMISF force reconstruction. The improved state-space system obtained using the
DMISF becomes:
{ } [ ]{ ̇ } [ ][ ]{ ̇ }
[ ] [[ ] [ ] [ ]] (2.2.5.4)
34
{ ̈ } [ ]{ ̇ } [ ]{ ̇ }
In equation (2.2.5.4) [ ] is a matrix that isolates the input sub vector { ̇ } out of the input
vector { ̇ }, in conformity with the definition of the state equation in (2.2.5.1) and (2.2.5.2).
The inverse structural filter procedure in (2.2.4.3) and (2.2.4.4) is then applied to the state-
space system in (2.2.5.4) by:
{ } [ ̂ ]{ ̇ } [ ̂ ]{ ̈ }
{ ̇ } [ ̂ ]{ ̇ } [ ̂ ]{ ̈ } .. (2.2.5.5)
where,
̂ ̂
̂ ̂ (2.2.5.6)
This inverse system estimates the time derivate of the applied loads at time instants between
to from the responses between to . If response data is recorded until a given time
instant , the above displayed inverse state-space system allows multiple load estimates.
DMISF has shown in [12] and [26] to be more stable and slightly more accurate than ISF,
SWAT and FD methods.
35
3 NUMERICAL METHODS
(3.1.1.1)
[ ] (3.1.1.2)
36
Therefore one can conclude that the disk element contributes with a mass matrix (the first in
equation (3.1.2.1)) and with a gyroscopic matrix which is the one that depend on the rotating speed .
This element corresponds to the MASS21 element in ANSYS®.
The shaft
The shaft element of length will be modeled as a two node beam element with constant
circular cross-section, and is illustrated in the Figure 3.1. The Euler-Bernoulli beam theory will be the
considered for the elastic shaft since the modeled shaft is long and slender. Other theories, like the
Timoshenko beam theory is proper for shorter beams and larger cross-sections, and includes the
influence cause by shear loads on the shaft. This affects its total stiffness. Since this effect could be
important in rotor shafts, a way to include a shear correction is presented.
Both nodes are characterized by the displacement vector defined in equation (3.1.1.2). The
nodal displacement vector for the shaft element can be written as:
[ ] (3.1.2.2)
Now in order to build the shaft finite element the displacements must be related to shape
functions ( ), as the next equation states:
( )
( ) (3.1.2.3)
where and include the displacements contained in (3.1.2.2) in the and directions
respectively. Concretizing this last statement it results:
[ ]
[ ] (3.1.2.4)
37
The shape functions in (3.1.2.3) ( ) and ( ) will be those typically used in a beam in
bending, and they are:
( ) [ ]
( ) [ ]
(3.1.2.5)
Introducing the equations in (3.1.2.3) and its derivatives in the total kinetic energy expression
(2.1.1.5), computing the integrations and introducing the resulting expression on the Lagrange
equation, it results:
( ̇ ) ( ) ̈ ̇ (3.1.2.6)
(3.1.2.7)
[ ]
(3.1.2.8)
[ ]
38
(3.1.2.9)
[ ]
Likewise, introducing again the equations in (3.1.2.3) in the strain energy expression in
(2.1.1.6), performing all the integral operations involved, and applying the Lagrange equation, it yields:
( ) (3.1.2.10)
( ) ( )
( ) ( )
( )
( ) ( )
[ ( ) ( ) ]
(3.1.2.11)
(3.1.2.12)
[ ]
where is the correction factor that takes into account the shear effect in the shafts stiffness, and it is
equal to:
39
(3.1.2.13)
(3.1.2.14)
( )
where is the Poisson’s ratio of the shaft’s material, and is the reduced area of its cross-section
approximately equal to the cross-section area . The matrix represents the classical stiffness matrix
of the shaft while the matrix represents the stiffness due to application of an axial force on the
shaft.
Though based in the Timoshenko theory the ANSYS® element BEAM188 (2 node/6 dofs), is
the assigned element to model the shaft. The determinant reason to choose this element was the fact
that this is the only two node beam element available in ANSYS® that includes the Coriolis effect in its
computations. Besides that, it is recommended for slender to moderately slender beams, such the
ones considered to model the shafts of the rotors in this work.
The Bearings
The bearings in the finite element system are subjected to the same assumptions referred on
the chapter 2.1.1. Adapting equation (2.1.1.9) to the displacement vector considered in the finite
element model in (3.1.1.2), it results for the forces acting on the bearings:
̇
̇
[ ] [ ][ ] [ ][ ̇ ] (3.1.2.15)
̇
In ANSYS® this was modeled using the spring/damper element COMBIN14
Mass unbalance
The kinetic energy related to a mass placed with a specific offset in relation to the rotation axis
is shown in equation (2.1.1.12). Like has been done before if this kinetic energy expression is
introduced in the Lagrange equation the finite element model of this mass gives:
( ̇) [ ] (3.1.2.16)
As it is shown in equation 3.1.2.16 the mass unbalance effect acts along the two translational
directions stated on the displacement vector in equation 3.1.1.2. The way the mass unbalance is
implemented in ANSYS® will be subject of discussion on the chapter 3.3.
40
to give the natural frequencies and the modal vectors. Now in this section the exact procedure of the
solution is made clear, so that the numerical implementation become fully comprehended.
Two methods of solving this problem will be very shortly presented: The pseudo-modal
method, quicker and more efficient, and the direct method, more precise but more demanding of
computational effort.
The general equation used to compute the natural frequencies and the modal vectors is:
[ ] ̈ [ ] ̇ [ ] (3.1.3.1)
where is the modal displacement and [ ], [ ] and [ ] are the mass, damping and stiffness matrices
respectively.
The pseudo-modal method
The pseudo-modal method starts with the solution of the eigenvectors of the following reduced
system:
[ ] ̈ [ ] (3.1.3.2)
{ } (3.1.3.3)
{ } [ ]{ } ̈ { } [ ]{ } ̇ { } [ ]{ }
(3.1.3.4)
In rotors, as it was seen, the damping matrix depends on the rotating speed , therefore
equation (3.1.3.4) should be solved for specific values of .
The solutions are sought as before:
(3.1.3.5)
[ ] (3.1.3.6)
(3.1.3.7)
41
Manipulating equation (3.1.3.6) it results:
[ ][ ] [ ] (3.1.3.8)
⏟
Solving the Eigenproblem of matrix in (3.1.3.8), the modal vectors are extracted from the
semi-vector and the resulting eigenvalues are included in in the form already defined in the
beginning of the section 2.1.3.
The direct method
Direct method corresponds to the solution of the equation (3.1.3.1) without intermediate steps.
There are several numeric functions capable of that. In Matlab®, for instance, this is possible using
function polyeig that solves the polynomial eigenvalue/eigenvector problem of second degree stated
below:
[ ]{ } { } (3.1.3.9)
if
42
are employed as described in 2.2.4.-2.2.5. After the full transient analysis is performed, the
accelerations at the desire dofs are extracted using the ANSYS® time history postprocessor.
These measurements are then loaded in Matlab® where the response vector is built. The way
the measurements are organized inside the vector depend on the sequence of the dofs of the
numerical method used to build the inverse system of the structure.
After this is accomplished, the Matlab® code proceeds with the specific force identification
algorithms for each of the implemented time-domain force identification methods.
{ ( )} [
⏟ [ ] [ ] [ ]] { ( )} (3.2.2.1)
[ ( )]
This is done in an arbitrary range of frequencies chosen by the user.
The next step is to choose what dofs responses should be included either in vector { } and
{ }, this can be also arbitrated by the user.
The code is programmed to scroll all the possible combinations of application points of the
force, and to successively apply equations (2.2.1.10-11). In each combination step the program gets
the proper values for the matrices [ ] and [ ] from the system’s [ ( )] matrix that was derived
using the finite element model in the chosen frequency interval.
By definition the code will get the nodes where the lowest error was registered and considers
that the loads are acting in the resulting nodes. Finally the amplitude of the loads is then computed
with equation (2.2.1.7).
3.2.3 SWAT
In this method the first step consists of determining the weighting matrix as described in
equation 2.2.2.4. For that the elastic modal vectors of the considered finite element model of the
structure are first determined through the solution of the respectively associated
eigenvalue/eigenvector problem using the direct method. This second finite element model must be as
close as possible to the ANSYS® finite element model where responses are extracted, i.e. with similar
element and equal discretization of the mesh, and it is built in a separate code. Then the rigid body
modes vectors are computed in conformity with the structural system under study. The user can
choose what rigid body vectors to include, since the SWAT method, as described in section 2.2.2
estimates the force in the direction of the correspondent rigid body motions.
To show this let us consider, for instance, a beam finite element. This beam element is
characterized by three dofs per node, so it has also three rigid body modes. Two of them are
translational (longitudinal translation is the first nodal dof and vertical translation the second), and a
third one is rotational (rotation about an axis perpendicular to the beam’s lateral section, which is the
also the third nodal dof). Figure 6.8 in Appendix 5 shows the mentioned beam finite element. In a
43
translational rigid body mode vector, the vertical for instance, every single dof in that direction moves
and the others have a null modal displacement. Generalizing for the other two rigid body modes, one
can conclude that for the beam element the rigid body vectors are:
[ ] [ ]
[ ] [ ] (3.2.3.1)
[ ] [ ]
Thus, one can conclude that in the developed code, the rigid body vectors are built by making
all the dofs in the direction in which the rigid body motion is being considered equal to one, while the
other vector positions are considered null.
As described, the elastic mode vectors are obtained solving the eigenvector problem, which in
a generic dynamic system is:
[ ]{ } { } (3.2.3.2)
with:
√
Here , and are the mass, damping and stiffness matrices of the studied system. And
the eigenvalues of the structure. In the case these { } are complex, only the real part is considered
as described in [12]. The number of flexible modes to include in the weighting matrix computation is
chosen by the user.
Now the weighting matrix is computable and the programmed Matlab® follows the sequence
already explained in chapter 2.2.2. As is visible in equation 2.2.2.1, the weighting matrix [ ] is applied
to the vector { } which includes the instant accelerations in every dof, in order to compute the
rigid body accelerations.
The extracted responses for a specific time instant will be placed in the vector { } attending
the sequence in the modal vectors of the structure.
Then, equation 2.2.2.5 is applied to determine the load vector in the considered instant, where
the inertial characteristics are immediately determined, since the geometry and the density of the
material is assumed to be known.
Finally a cycle is applied so that equations 2.2.2.2 and 2.2.2.5 are successively applied in the
considered time domain.
3.2.4 ISF
The ISF method begins with the definition of the state-space system matrices. As we saw
back on chapter 2.2.3 there’s actually two ways of doing so: through the system’s matrices, or through
modal parameters.
Only the sate space system based on the system matrices was numerically implemented. The
programmed code follows the explanation shown in chapter 2.2.3. The system matrices were built in a
44
separate finite element program, as was previously said for the SWAT method this finite element
model is similar to the finite element model from where the structure’s responses were extracted. The
user must also introduce matrix [ ], since it will indicate the location of the degree-of-freedom where
the forces are to be reconstructed, by introducing 1 in the array position(s) corresponding to the force
location and 0 in the remaining array positions.
The next step consists on defining the state-space system itself. This was done using the
Simulink® tools present in Matlab®. This tools are very important, since they include a function ‘s2d’
that automatically discretizes the state-space system giving its altered matrices. All the user needs to
introduce in the programmed is the constant time step used on the discretization. Note that the used
time step must be equal to the acquisition interval used in the transient analysis made on ANSYS®
from where the responses were extracted. So far the described procedure is shared between the ISF
and the DMISF techniques.
Before the inverse system is programmed, the non-causal lead must be introduced by the
user along with the number of Markov parameters he wants to use.
Finally the inverse system is ready for computation. The inverse system matrices must first be
built according to the equations (2.2.4.4). After this, the Markov parameters of the inverse can be
computed according to (2.2.4.7), at last equation (2.2.4.10) is cyclically applied to each time step
between and , being the total number of sub-steps of the extracted response.
Since only the first time derivative is obtained from equation (2.2.4.10), a numeric integration
method must be applied in order to obtain the force reconstruction. This aspect is common both to ISF
and DMISF methods.
The Newton-Cotes Simpson rule is the used numeric integration method, applied in the
context of the ISF method, and noticing that the time step is constant, it results:
[ ̇ ̇ ̇ ] (3.2.4.1)
where represents the time step number. The integration is applied between time step until time
step .
In the case a force impulse rather than a harmonic force was estimated, the Simpson rule has
proven to lead to wrong estimations. Therefore, for the examples where the force impulse is being
estimated, the trapezoidal integration method is used, following the equation (3.2.4.2) displayed just
below.
̇ ̇
[ ] ( ) (3.2.4.2)
3.2.5 DMISF
The first steps of the DMISF are common to the ISF method as it was mentioned above. After
the discrete space-state system matrices are defined, the user must introduce the delay he intends
to use to run the force reconstruction. For the chosen delay the code is now ready to build the
matrices [ ] and [ ] in equation (2.2.5.3), using the computed discrete space-state matrices.
45
The next step consists of obtaining the inverse space-state system following the procedure
and matricidal transformations described in equations (2.2.5.5-6). Vector { ̇ } is then obtained via
equation (2.2.5.5), which contains reconstructions of the load first time derivative. Only the last line
of this vector is considered for integration, i.e. the positions corresponding to { ̇ }, since they
correspond to the last and thus more perfected, force estimation.
Therefore the following integration necessary for the force reconstruction gives, according to
Simpson:
[ ̇ ̇ ̇ ]
(3.2.5.1)
where represent once more the time step. The integration is applied between time step until
time step .
Again for the estimation of force impulses the trapezoidal integration method was employed as
follows:
̇ ̇
[ ] ( ) (3.2.5.2)
In this work only the vibrations caused by the presence of an unbalance mass in the disk of
the rotor are analyzed.
As demonstrated through equation 2.1.3.5 the force associated with a mass unbalance can be
described by two force components along the rotor’s generalized coordinates and dubbed
and respectively. This is illustrated in Figure 3.2.
Figure 3.2 – Force components due to the presence of a mass unbalance in the disk of the rotor.
In ANSYS® this was modeled by applying to punctual forces along the two directions
perpendicular to the rotating axis. The way these two components are defined depends on the
performed analysis.
46
Knowing that the forces related with mass unbalances of the disk are synchronous, one just
needs to specify the force amplitude and force component phase angle when performing a harmonic
analysis in ANSYS®.
When a transient analysis is performed there is no way to define a synchronous force
automatically. So it is necessary to compute the force components in each time instant and save them
in specific arrays in coherence with the force equations 2.1.3.5 associated with the mass unbalance.
The program will go through the saved arrays and apply the correspondent loads in the proper time
instants.
47
4 APPLICATIONS AND DISCUSSION
48
4.1.1 The Finite Element Model
The modeling in ANSYS® of the considered problem is based on the examples of the
Rotordynamics chapter of the ANSYS® Manual. Here the shafts are modeled using a set of equally
spaced axial nodes, to which are attributed beam elements (BEAM 188 with quadratic polynomials)
with the section properties of each respective shaft. In this model the inner shaft is divided in 15 equal
parts (16 nodes) and the outer is divided in 4 equal parts (5 nodes).
The disks are modeled as punctual masses (MASS21), with the respective inertial properties,
in the nodes corresponding to their location and to their shaft.
The bearings are modeled using two spring-damper elements (COMBIN14), and are
connected to the node in the bearing location corresponding to the outer shaft.
The outer and the inner shafts are considered connected in the right end node of the outer
one. Here a constrain linking the displacements in the and directions between the inner and the
outer shafts nodes in this location is applied.
The inner shaft is simple supported in both ends so the displacements in the corresponding
nodes are constrained. The outer one is simple supported only in the left end node, meaning therefore
that the displacements in the corresponding node are subjected to the same constrains.
Additionally, the displacements and rotations about the axis are constrained in all nodes.
The only exception is in the case where an axial moment is applied in the inner shaft. Here the
rotations about are only constrained in the nodes of the outer shaft and in the right end node of the
inner shaft.
The net change of the values of the critical speeds will work as convergence criteria in this
validation.
The elements that were used in these two additional meshes were the same used for each
rotor component as it was stated on chapter 4.1.1. In the coarser mesh was employed the half of the
elements used in the previous described mesh. In the most refined mesh was employed the double of
the elements per shaft used in the mesh described on chapter 4.1.1.
Figure 4.2 along with table 4.1 show the Campbell diagrams and the values of the inner and
outer shaft synchronous critical speeds for the three studied meshes.
49
Coarse (7/2) Used (15/4) Refined (30/8)
Figure 4.2 – Campbell Diagram convergence – the brackets above represent: (number of elements in the inner
shaft/number of elements in the outer shaft)
Table 4.1 – Critical speeds of each mesh and the correspondent errors computed using the FE solution as
reference
Coarse Used Refined
Inner synch Critical Speeds (rpm)
FW BW FW BW FW BW
Inner Synch 12339 14410 8710.453 11406.598 8710.451 11406.595
Outer Synch 2611 2693 1908.433 2011.715 1908.433 2011.714
Error coarse/used Error used/refined
Errors (%)
FW BW FW BW
Inner Synch 41.7 26.3 2.296e-5 2.630e-5
Outer Synch 36.8 33.9 0.000 4.97e-5
From the mentioned Figure 4.2 and table 4.1 it is visible that the error between the used finite
element model and the most refined mesh is irrelevant (in the order of %). So the quality of the
used discretization is confirmed for the purpose of the natural frequencies calculation.
The bearing properties and the shafts’ rotating speeds will differ from each particular solution.
50
4.1.4 Results and Results Discussion
It will be now presented the Campbell and Response Diagrams for various conditions of
elastic bearing properties, like stiffness asymmetry and damping. The influence of multiple inner and
outer rotating speed ratios, which relation is governed by the parameter that was first presented in
equation (2.1.2.17) will be also a focus of the following section. The effects caused by a torsional
moment placed in the inner shaft will also be considered late on this section.
The Campbell Diagrams will be drawn taking into consideration a step of between
each computation of the natural frequencies. In the Response Diagrams it is used a step of
between each response measurement. The term rotating speed in the diagrams refers to the rotating
speed of the inmost shaft, i.e. the value of . The Campbell Diagrams include the representation of
the inner shaft synchronous ( ) and the outer shaft synchronous ( ) lines.
The errors between the Rayleigh-Ritz and finite element methods presented in this section are
always computed relatively to the finite element numeric value.
Symmetric system
For the symmetric system the elastic bearing properties are:
Example 1
For :
Table 4.2 – Critical speeds associated with the previous Campbell diagram – RR/FE error
Inner shaft synchronous Outer shaft synchronous
Critical Speeds
(rpm) (rpm)
FW BW FW BW
Rayleigh-Ritz 8743 12008 1940 2064
Finite Element 8710 11407 1908 2012
Error (%) 0.38 5.27 1.68 2.58
51
There are some remarkable aspects in Figure 4.3, namely:
1. The obtained curves using the finite element method have lower natural
frequencies comparing with the computed solutions using the Rayleigh-Ritz method. This result
has already been documented in the literature, particularly in [4]. Being the shape equation used in the
Rayleigh-Ritz method the first mode exact solution of a beam in bending, and the finite element
method an approximation method, it was expectable that the opposite was verified, since the exact
solution’s shape function would adapt better to the first mode shape. To explain this disparity let us
analyze the first mode shape. The rotor’s shaft shape in bending is deformed by the presence of the
disk, since it concentrates the majority of the rotor’s weight. The interaction with the bearings would
also play a role if the tested rotor had one. The modal shape obtained with the finite element method
therefore presents lower natural frequencies because the used mesh allows a better agreement
between the deformation computed numerically and the deformation that is actually being observed in
the shaft. To prove this way of reasoning, a detailed study of the influence of the disk’s weight and
position is present in the Appendix 3 to this work.
2. The BW natural frequencies grow and the FW natural frequencies decrease with
increasing rotating speed.. The gyroscopic effect is rather complex in a Multirotor configuration. In
this particular case , this means that the outer shaft is rotating 5 times faster and in the
opposite direction relative to the inner shaft. Consequently the skew-symmetric part of the damping
matrix governed by the term is negative. This means therefore that the gyroscopic effect will
act in the opposite sense it used to if the Multirotor was constituted solely by the inner shaft, reducing
the stiffness whenever it is in FW and increasing it when the Multirotor is in BW.
Figure 4.4 – Finite element curves make use of the direct method (chapter 3.1.3)
52
1. The Amplitude grows, and approaches asymptotically the inner shaft
synchronous FW critical rotating speed of the considered symmetric system, after which it
decreases and stabilizes. The reason why no resonance is verified at the BW critical rotating speed
is related with the fact that the rotor is symmetric. Consequently the rotor’s orbits will be circular.
Moreover since the unbalance is present in disk 1, the excitation force is synchronous and co-
rotational with the inner shaft who governs the overall rotor orbit. In this particular situation as it was
justified in chapter 2.1.3 there is only one resonance operating point, and it corresponds to the FW
mode, since it is being excited by a co-rotational excitation force.
2. The finite element results differ from those obtained with the Raleigh-Ritz
method. The asymptote of the finite element computed response is verified for slightly lower rotating
speeds since, as it was previously explained, the computed natural frequencies are lower using the
finite element method than using the Rayleigh-Ritz method. Moreover the absolute values of the
amplitude obtained with each of the employed methods differ slightly from each other. Again this has
to be with the fact that the finite element method results adapt better to the modal shape of the
Multirotor first bending mode (see Appendix 3).
Figure 4.5 – Finite element curves make use of the direct method (chapter 3.1.3)
53
critical speed in the Campbell Diagram, i.e. . Additionally, close to the , another
asymptote is visible. This happens because, as it was said in the ‘Solving Methods’ in the beginning of
chapter 4.1, the harmonic solution of ANSYS® uses the direct method to compute the natural
frequencies and the pseudo-modal method was the employed method for the natural frequencies
computation to construct the Campbell diagrams on this work. As it is shown in the Appendix 4 to this
work, if the direct method was used to determine the natural frequencies and consequently to build the
Campbell Diagrams, the frequency curves would present a deviation from the solutions obtained using
the pseudo-modal method, especially at higher frequencies. This explains the deviation in the finite
element results. Additionally the direct method used in the harmonic analysis includes all the modes in
the frequency range under study, and not just the first mode as the Rayleigh-Ritz equations in chapter
2.1.3 contemplate. This explains the extra asymptote, which corresponds to the BW of the second
vibration mode of the symmetric Multirotor which is within the range of the harmonic analysis done for
the mass unbalance in disk 2.
3. The amplitude differences between the response curve obtained with the
Rayleigh-Ritz method and the curve obtained with the finite element method are larger for a
mass unbalance in disk 2 than for a mass unbalance in disk 1. This could be related with the fact
that when the Rayleigh.-Ritz method was deduced for the Multirotor on chapter 2.1.2, the outer shaft
was considered as rigid. This assumption was no longer taken into account when the finite element
model was built.
Example 2
For
54
Table 4.3 – Critical speeds associated with the previous Campbell diagram – RR/FE error
Critical Speeds Inner/outer shaft synchronous
(rpm)
Rayleigh-Ritz BW FW
9658 10371
Finite Element 9436 10188
Error (%) 2.35 1.80
Figure 4.7 – Finite element curves make use of the direct method (chapter 3.1.3)
55
Response diagram – mass unbalance in disk 2
Figure 4.8 – Finite element curves make use of the direct method (chapter 3.1.3)
Figures 4.7-8 confirm what was previously said about the effect of the mass unbalance in the
disks of the rotor. The differences between the finite elements and the Rayleigh-Ritz theories were
already been justified.
Example 3
For
56
Table 4.4 – Critical speeds associated with the previous Campbell diagram – RR/FE error
Inner shaft synchronous Outer shaft synchronous
Critical Speeds
(rpm) (rpm)
BW FW BW FW
Rayleigh-Ritz 7809 16624 1883 2140
Finite Element 7774 15223 1853 2084
Error (%) 0.45 9.20 1.62 2.69
In Figure 4.9 the only new aspect worth discussion is the large gap between the FW and BW
curves. This is easily explained by the co-rotate operation of both shafts in this case. This will result in
the largest absolute value of in the set of results presented until now. Consequently the
gyroscopic effect is significant which will lead to a sharper increase in the distance between the
curves.
Figure 4.10 – Finite element curves make use of the direct method (chapter 3.1.3)
57
Response diagram – mass unbalance in disk 2
Figure 4.11 – Finite element curves make use of the direct method (chapter 3.1.3)
The deviation between the finite elements calculated critical speeds, using the direct and
pseudo-modal methods, was already subject to discussion. Anyway one should notice the increased
deviation between the finite element FW critical speed computed in the Campbell diagram in Figure
4.9 ( ), the same critical speed compute with the Rayleigh-Ritz method ( )and the
correspondent asymptote location in the response diagram 4.10 ( ). This could be related
either with numeric aspects of the finite element solution of the direct method, or with errors introduced
by the assumptions taken in the Rayleigh-Ritz and the finite element pseudo-modal approximate
methods. Recall that the Rayleigh-Ritz method, for instance, doesn’t include a correction for the disk’s
inertia influence on the mode’s shape.
Asymmetric system
For the asymmetric system the elastic bearing properties are:
58
For :
Table 4.5 – Critical speeds associated with the previous Campbell diagram – RR/FE error
Inner shaft synchronous
Critical Speeds
(rpm)
FW BW
Rayleigh-Ritz
9224 13039
Finite Element 9180 12177
Error (%) 0.48 7.08
Comparing the Campbell Figure 4.12 with Figure 4.3, built in the same rotating speed ratio
conditions in the symmetric system, there are some noticeable changes.
1. When the rotor is stopped there are two natural frequencies instead of one.
Since the bearing stiffness characteristics are now asymmetric, the bending natural frequencies will
now be sensible to the different stiffness properties in the and directions. The same is to say,
that the characteristic equation (2.1.3.4) will now have two solutions for .
2. The FW critical speeds are higher for the asymmetric Multirotor than for the
symmetric and the BW critical speeds are lower. We can conclude that the asymmetric stiffness
properties of the bearing causes the FW and BW curves to drift apart relative to the symmetric
Multirotor FW and BW curves.
59
Response diagram – mass unbalance in disk 1
Figure 4.13 – Finite element curves make use of the direct method (chapter 3.1.3)
In Figure 4.13 is interesting to note that when asymmetric stiffness properties in the bearings
are verified, a FW excitation force such as that caused by a mass unbalance in disk 1, can also cause
resonant behavior at BW critical speeds.
The response diagram differences caused by a mass unbalance in disk 2 relative to a mass
unbalance in disk 1 were already widely discussed in the previous results. For this reason we will skip
the response diagram for a mass unbalance in disk 2 from this point onwards.
Damped system
The previous results were sufficient to show the most relevant differences between the finite
element and the Raleigh-Ritz method. For this reason, in the damped system we will just consider the
solutions obtained via finite element method, again using the pseudo-modal method for the Campbell
diagrams and the direct method for the response diagrams.
For the damped system the elastic bearing properties are:
60
Example 1
For :
Table 4.6 – Critical speeds associated with the previous Campbell diagram
Inner shaft synchronous
(rpm)
FW BW
9453 12153
The introduction of small amount of damping doesn’t change significantly the Campbell
diagram and the value of the critical speeds. This is visible if we compare the Figure 4.14 with Figure
4.12.
Response diagram – mass unbalance in disk 1
Figure 4.15 – Results make use of the direct method (chapter 3.1.3)
61
The Response diagram by the contrary shows substantial differences after the introduction of
damping. Asymptotes are now longer visible and the overall vibration amplitude has decreased.
Example 2
For
Table 4.7 – Critical speeds associated with the previous Campbell diagram
Inner shaft synchronous
(rpm)
BW FW
9186 9837
The Campbell diagram shown in Figure 4.16, presents major alterations in its curves with
increasing damping. High damping cause the FW whirl curve to appear above the BW curve for
lower rotating speeds even when , and for higher rotating speeds the curves’ relative
position switches, resulting in a rather complex evolution
62
Response diagram – mass unbalance in disk 1
Figure 4.17 – Results make use of the direct method (chapter 3.1.3)
Figure 4.17 shows once more the general tendency of the response diagram of an
increasingly damped rotor: Lower amplitude and less sharp peaks.
The Problem
Let us then start by reconstruction a force acting transversally to a beam in a previously known
location. The problem to be solved here is based on the beam that was subject of experimental
studies on the article of Allen and Carne about DMISF [12]. Figure 4.18 shows the beam in question.
In all examples the beam is considered free of any constrains.
63
Figure 4.18 – Beam Model with Force Impulse position represented by vector F
An impulse force is applied in the left-most end of the beam. In the preceding analysis it will be
tested the capacity of each time-domain method to reconstruct the force impulse.
Solving Methods
Every time-domain, namely the SWAT, ISF and DMISF methods will be applied to the
previous described beam following the numeric implementation explained on chapter 3.2, and the
fundamentals on chapter 2.2. The Frequency domain method was not implemented, because it was
only programmed for harmonic forces and not for impulse forces such as the one applied on this
example. An application of the FD method will be illustrated on the chapter 4.2.2.
The structure responses are extracted as mentioned on chapter 3.2.1. In the particular case of
the beam the translational and rotational accelerations in every node are extracted and used in the
reconstruction systems of each force reconstruction method.
The Finite Element Model
As was made clear on chapter 3, there are two finite element models. One for extract the
structure’s responses and another to build the systems matrices , and used in SWAT to
compute the flexible mode’s modal vectors and in the ISF and DMISF to build the state-space system
matrices. These finite element models have in common the discretization of the mesh. In the particular
case of this beam, 12 equally spaced beam elements were used as shown on Figure 4.19 along with
the finite element assembly table.
64
Table 4.8 – Assembly between global and local dofs of the Beam mesh
Element Number Element Type
1 Beam 1 2 3 4 5 6
2 Beam 4 5 6 7 8 9
3 Beam 7 8 9 10 11 12
4 Beam 10 11 12 13 14 15
5 Beam 13 14 15 16 17 18
6 Beam 16 17 18 19 20 21
7 Beam 19 20 21 22 23 24
8 Beam 22 23 24 25 26 27
9 Beam 25 26 27 28 29 30
10 Beam 28 29 30 31 32- 33-
11 Beam 31 32- 33- 34 35 36
12 Beam 34 35 36 37 38 39
In the ANSYS® finite element model, where the transient analysis of the beam subject to an
impulse was made, the beam model was built with BEAM188 (six degree-of-freedom per node)
elements, that makes use of quadratic polynomials. Though based in the Timoshenko beam theory it
is the recommended beam element for slender to moderately slender beams. The systems matrices
were built using a three degree-of-freedom per node beam element, whose element matrices and
characterization is stated on the Appendix 5 to this work.
To complete the beam model, it will be supported by two soft springs to simulate free-free
boundary conditions and the force will be applied on the leftmost node of the beam. In ANSYS® these
springs were introduced with COMBIN14 elements, while in the system’s matrices the springs’
stiffness was simply added in the proper positions of the system’s matrix.
Additionally a small value of internal damping is assigned to the beam to simulate the
experimental conditioning of the beam. This was done by considering a stiffness dependent damping
matrix, expressed by a positive integer factor , i.e. [ ] [ ] . Of course the spring stiffness is
not considered in the computation that’s why the designation is used to distinguish from the
general system’s stiffness matrix . In ANSYS® factor was introduced as a material property using
the ‘MP, DAMP’ function.
⁄
⁄
(Density)
(Young Modulus)
65
The Force input signal has the characteristics displayed in table 4.2.1.2.
Table 4.10 – Transient analysis and force impulse data
Recorded Time
Force Impulse Amplitude
Force Impulse Application Instant
Force Impulse Duration
Sampling Time Interval
Notice that the force impulse estimated by the SWAT method is about lower than the
force impulse actually acting in the beam. This is slightly higher than the error attributed to the
SWAT method in [12].
In this analysis there are several sources of error.
First of all the finite element model used to build the force reconstruction system have slightly
differences from the finite element model from where the responses were extracted. This means that
the force was estimated in a structural system with responses that not exactly match the response it
would have if the force impulse was applied on it.
66
Additionally in the mentioned work of Carne et al. the method was applied experimentally and
not numerically, which implies a different procedure in the way the elastic modes are obtained and
treated.
To conclude, it can be said that the SWAT method presents rather good results for the
estimation of the force impulse amplitude in the beam. After the force impulse is removed the SWAT
method stills estimating a non-zero force value, but even though the main estimated curve peak is
clearly located between the time intervals in which the force impulse was acting in the beam.
2. ISF
For the ISF method the lower delay parameter value at which stable results were obtained
was , the chosen number of Markov parameters to use in this analysis was . The force is being
estimated in the dof where the force is applied. It corresponds to dof number 2 as shown on Figure
4.2.1.1. So vector [ ], has a 1 in the second array position and 0 in the remaining. The results thus
obtained are displayed in Figure 4.21. Again with the entire time domain in the left-hand side and the
first time instants displayed on the right-hand side.
As expected from the results in the literature [12], the ISF leads to very good estimations with
low errors. The estimated curve practically overlays the measured force impulse. Despite this fact, it is
noticeable the residual ringing conducted by the estimated force curve after the force impulse is
removed.
3. DMISF
The first stable solution for the beam using DMISF was obtained for . As in the ISF
method vector [ ] will make the force estimation to take place in the dof where we know the force is
applied. Figure 4.22 shows the obtained results for the entire time domain on the left-hand side and
the first time instants ‘zoom’ on the right-hand side.
67
Figure 4.22 – DMISF applied to the beam
The results obtained with DMISF are as good as that obtained with the ISF method. The
estimated force almost overlaps the measured force in the impulse application time instants. Despite
this, it is noticeable a slight reduction in the estimation curve ringing in the post-force impulse time
period.
This set of results globally suggests that the DMISF presents better results than the other
methods, since the estimated force follows in perfection the measured force and it has the less
significant residual ringing.
The Problem
Now let us consider a symmetric rotor, such as that shown in Figure 4.23 with a simple
supported shaft, no elastic bearings, and one disk.
68
Solving Methods
Firstly the Frequency domain method will be applied to the current problem in order to localize
the input load. Then the SWAT, ISF and DMISF methods will be successively applied to the rotor to
determine the time-history of the force. The method implementation follows the description made on
chapters 3.2 and 2.2.
Once more time the response is extracted from a finite element model as mentioned on
chapter 3.2.1. The translational and rotation accelerations about the two perpendicular axes to the
rotor rotating axis, in every model node, compose the response data from which forces are to be
estimated.
The Finite Element Model
Right below in Figure 4.24, the used mesh in the finite element models is represented for this
rotor along with the assembly matrix used to build the matrices of the system, with every dof number
identified.
Figure 4.24 – Finite element model with the element numbers in black and the nodal numbers in blue
Table 4.11 – Assembly between global and local dofs of the Monorotor mesh
Element Number Element Type
1 Beam 1 2 3 4 5 6 7 8
2 Beam 5 6 7 8 9 10 11 12
3 Beam 9 10 11 12 13 14 15 16
4 Beam 13 14 15 16 17 18 19 20
5 Beam 17 18 19 20 21 22 23 24
6 Beam 21 22 23 24 25 26 27 28
7 Beam 25 26 27 28 29 30 31 32
8 Beam 29 30 31 32 33 34 35 36
9 Beam 33 34 35 36 37 38 39 40
10 Punctual Mass 13 14 15 16 - - - -
The element matrices for the rotor components are shown on chapter 3.1.2. In this example
the used element stiffness matrices are the ones which include a shear correction. Since no axial
force is applied in the rotor, the axial force stiffness influence is null.
In the ANSYS® finite element model the shaft is modeled with BEAM188 (six dof per node)
elements with quadratic polynomials, the disk is represented by a punctual mass of equivalent inertia
69
using the element MASS21, and the displacements in the transversal directions are constrained at the
shaft’s extremity nodes.
There’s no addition of damping factors.
Numeric Data and Force Input Time-history
Table 4.12 includes all its relevant material and dimensional parameters.
Table 4.12 – Numeric data of the Monorotor
Material Constants Dimensions Mass Unbalance
(Density)
(Poisson Ratio)
Table 4.13 includes all relevant data used in the frequency domain analysis.
Table 4.13 – Relevant data for the frequency domain force identification method
Initial frequency
Final frequency
Number of solutions
The time domain analysis parameters as well as the value of the rotating speed that influences
the force input signal has the characteristics displayed in Table 4.14.
Table 4.14 – Relevant data for the time domain force identification methods
Recorded Time
Rotating speed
Sampling Time Interval
70
Appling the Frequency domain method as stated in the parameters present in table 4.13, the
error graphic for each force combination, computed as demonstrated in equation (2.2.1.11), it results a
graphic such as Figure 4.25 displays below.
As is perceptible in Figure 4.25, the error is minimized in the force combination. This
combination corresponds to the three forces acting respectively in the nodes number , and , i.e.
in the rigid supports and in the disk’s node.
The force position is therefore determined as being placed at the disk.
2. SWAT
The rotor is now rotating at constant speed and it is intended to determine the force time
history of the applied load at the disk. Recall that the estimated force determined with the SWAT
method estimates the resultant force at the structure’s gravity center. As the disk is rotating, so does
the force, since the mass unbalance that causes the unbalance rotates with it. Therefore the force will
be reconstructed in both rotor transversal directions. The following results were obtained using 20
elastic modes, and the two translation rigid body mode vectors (each for nodal translation dof) are built
as stated on chapter 3.2.3.
71
The results for each direction are displayed in Figure 4.26 below.
72
In general the estimated force curve closely follows the measured force curve.
Despite of that it should be mentioned, the difficulty demonstrated by the estimated force
curve to follow the peaks at lower amplitudes, exceeding in some points the amplitude of the
measured force.
Also note the intense ringing of the estimated force in the first time instants of the
reconstructed force in the direction.
4. DMISF
Using the same [ ] vectors that were used in the ISF method in order to obtain an estimation
of the force in the and directions, and applying a delay parameter of , it results two
graphics. They are displayed in Figure 4.28.
Notice that the ring is much less intense in the DMISF estimated curve when in comparison
with the ISF estimated curve. Nevertheless, the estimated force amplitude especially in the higher
force peaks fails to meet a perfect matching with the measured force curve.
Globally it can be said that time-domain force identification methods present good results
when applied to simple examples of Rotordynamics. SWAT presents good results qualitatively
speaking, but fails to present consisting force amplitude reconstruction. The ISF method despite
showing some ringing has proven to be highly reliable in force amplitude prediction, while the DMISF
method has shown difficulties in determining the force amplitude at the higher force amplitude peaks,
but diminishes the ringing of the force curve. This is consistent with what was obtained in the beam
application example.
73
The DUPRIN project was launched with the support of the European Commission in 1990, in
which 14 European research partners were involved. They had the objective of developing a High
bypass, low fuel consumption and highly efficient jet engine to meet the growing needs of
sustainability in the aircraft propulsion market [5]. The achieved design included a compressed air fed
four stage turbine that supplied the shafts connected with two counter-rotating ducted propfans with
160kW allowing a maximum rotating speed of up to 16 000 rpm. [6]
Figure 4.29 – The DUPRIN Propfan in its test facility – source: [6]
The first task to perform will be the reproduction of the Campbell Diagram and the first two
modal, and to compare the achieved solutions with that present on the literature [6]. For that a self-
made finite element model in ANSYS® is used.
To finish the chapter, the ISF method will be applied to the finite model of a Propfan engine to
reconstruct a force whose location is known but whose amplitude and periodicity is ignored. This last
example intends to show that time- domain force identification can be of practical use in aircraft
propulsion applications.
74
In what concerns the materials, a titanium alloy was assumed to the elements corresponding
to the fans, and a stainless steel was assumed in the remaining elements representing the Propfan’s
shaft.
Lalanne et al. show a picture of their Propfan finite element model when they performed
studies in the subject [6]. This picture was used to scale the Propfan engine finite element model used
in this work. To define the dimensions of the model of the engine a cross-reference between the scale
and the two known dimensions was made determining therefore the remaining dimensions of the
Propfan, resulting in the schematic shown in Figure 4.30.
The Propfan model has three bearings whose longitudinal and radial coordinates are
displayed in Figure 4.30 as well, and they include only stiffness terms in the and components. A
material damping term was also considered, equivalent to a quality factor . This was performed
because the authors in [6] considered it appropriate to describe the real dynamic behavior of the
engine. The numeric values of the stiffness of the bearings were not provided as well, therefore its
values were adjusted until an acceptable agreement between the author’s results and that present in
[6] was met.
Solving Methods
In order to obtain the intended results, a finite element model was built in ANSYS® and a
modal analysis performed. After this the ANSYS® Rotordynamics Post-processing tools were used to
extract the Campbell diagram representation, to compute the correspondent critical speeds and to
display the orbit shapes associated with the first two vibration modes.
75
Additionally, and as it was performed for the Multirotor example in chapter 4.1, the disks were
modeled by punctual masses of equivalent inertia, using element MASS 21 and the bearings’ stiffness
was modeled with COMBIN14 spring elements.
To simplify the finite element model the stiffness of the spring connecting the inner and the
outer shaft was considered infinite. This was also an assumption taken in the Multirotor example
present in this work, and it has given satisfactory results.
The quality factor was was accounted for making use of the function MP/ DMPR.
The resultant finite element model is shown in Figure 4.31 just below.
⁄ ⁄
(Density) (Density)
76
Results and Results Discussion
Campbell Diagram
The obtained Campbell Diagram for the Propfan used in this work is shown in Figure 4.32
along with the Campbell Diagram obtained in [6] for the same structure. The associated critical speeds
are indicated in Table 4.17.
Table 4.17 – Error computed relative to the values obtained with the author’s finite element model
Inner/Outer shaft synchronous
Critical Speeds
(rpm)
FW BW
Finite Element
7146 8421
Lalanne et al. 7019 8357
Error (%) 1.78 0.76
Modal orbits
The modal orbits for the first two vibration modes are here depicted in Figure 4.33 and
compared with the results of Lalanne et al. This is done to confirm the coherence of the obtained
modal analysis solution.
77
Figure 4.33 – First two modal shapes of the Propfan; Author’s model above/Model in [6] below; FW (Left) and
BW(Right)
Several conclusions can be taken from the established comparisons evident in Figures 4.32-
33.
1. The dynamic behavior is coherent, i.e. in the frequency range contained in the
Campbell Diagrams in Figure 4.32 the whirl sequence at which the vibration modes appear is the
same in both finite element models. Adding to this the modal shapes associated with the first two
vibration modes computed with the developed finite element model coincide with the modal shapes
obtained by Lalanne et al. in their work about the DUPRIN project.
2. The evolution of the first two frequency curves as well as the critical speed
values is similar in both finite element models as Figure 4.32 and Table 4.17 state.
These two conclusions suggest the reliability of the developed finite element model and give
the necessary confidence to try the application of a time-domain force reconstruction method in this
Propfan model.
3. The verified differences for the third and fourth vibration modes should be related with
incorrect assumptions in the dimensioning and material constant choice for each Propfan part,
between the in-house developed Propfan finite element model and the one developed in the studies of
Lalanne et al. [6]
78
4.3.2 Reconstruction of a Force Caused by an Unbalance in the Fan
The Problem
The force excitation at which the Propfan shown in Figure 4.30 will be subjected is caused by
a mass unbalance in Fan 1. The Propfan will start from rest and will accelerate at constant rate to a
speed of 625 achieved 0.5 seconds afterwards. The two shafts constituting the Propfan are
running at the same speed but in opposite senses. The Fundaments Chapter 2.1 of this work didn’t
contain any mention to transient motion of rotors. Therefore a brief description of these motions is
provided in the Appendix 6 to this work.
Solving Methods
The finite element model presented in the last section (4.3.1) is slightly changed before being
applied to this problem. The couple between the two rotors used in the previous model to connect the
inner and the outer shafts was removed, and a small stiffness spring was introduced in both and
directions. This was performed because when the Transient analysis was carried out in the finite
element model of the Propfan, the force reported to be applied in the node of the FAN 1 differed from
the force actually being applied in that node, leading to erroneous values of the response. The author
could only overtake this issue, in the available time frame for this work, by removing the referred
coupling and by introducing the springs. This means that the modal analysis that was previously
undertaken not actually corresponds to the behavior of the finite element model employed in the
Transient analysis code.
A full transient analysis, in the previously described conditions, was performed in ANSYS® in
order to get the Propfan responses. The force was modeled as two punctual forces acting along the
and components as described on chapter 3.3, that follow the force functions present in equation
A6.8 for transient motion deduced in Appendix 6.
Since the Propfan model is much more complex than the Symmetric rotor considered in
chapter 4.2.2, the accelerations will be only registered in the translational dofs of the even nodes of
the inner rotor (13 nodes – 26 measurements), and in the translational dofs of each node of the outer
shaft (16 nodes – 32 measurements) limiting the response array in each time instant to 58 values.
Additionally in order to apply the ISF, the state-space system was built using the system’s
matrices. These matrices were built using the Substructure tool of ANSYS® using the same 58 dofs
from where the response was extracted as master dofs, i.e. ANSYS® built a simplified model from the
dofs the user wants to include in the force reconstruction process. ISF is now ready for application
according to chapters 2.2.4 and 3.2.4, with the nuance that the damping matrix [ ] that include the
gyroscopic terms is now varying from time instant to time instant, and that new stiffness matrix terms
appear.
This is achieved through cyclic computation.
One may think that the problem is simply solved by applying the ISF method with matrix
[ ] changing in each time instant. This however has led to unstable solutions. Another cyclic
methodology is proposed by the author.
79
Each cycle step will correspond to a time instant. Firstly the rotating speed is fixed with its
instantaneous value. Then the estimated force is computed for the time domain comprising the initial
time instant and the actual time instant with constant rotating speed equal to the instantaneous
rotating speed. When the force estimation is obtained, only the value corresponding to the step time
instant is saved to the force estimation array.
This methodology has presented stable solutions but increased computation time.
ISF will be the only force reconstruction method applied since it demonstrated to be the most
computer efficient method in the author’s experience.
Numerical Data
The data in Table 4.16 is again applicable to this case. It must be now added the Transient
analysis data, the mass unbalance data as well as the stiffness of the springs connecting the inner
shaft with the outer one, included in Table 4.18.
Table 4.18 – Transient analysis and mass unbalance data
Recorded Time
Sampling Time Interval
Mass unbalance
Offset
Spring stiffness data
The estimated curves present a reasonable agreement with the measured force at lower force
amplitudes and a considerable deviation at the higher amplitudes. These deviations may be related
with the reduced set of measurement points and with imprecisions in the finite element model
developed in ANSYS®, from which the responses were obtained.
Additionally the elapsed analysis time starts to become an issue. In the average of the two
reconstructions the code needed 647 seconds to reconstruct 0.5 seconds of force time-history. The
80
ISF method was used at the first place and the number of Markov parameters was kept low, since this
configuration has shown the most computer efficient performance.
81
5 CONCLUSIONS AND FURTHER DEVELOPMENTS
5.1 Conclusions
82
of the solution of ISF and DMISF and how the inclusion of more or less flexible modes would affect the
ability of SWAT to estimate better or worse the forces. It would be interesting though to develop this
kind of work in the future, for the sake of the optimization of the force identification process.
To improve the stability of ISF and DMISF methods, it would be interesting to employ non-
linear control optimization techniques as proposed in the Appendix of the paper by Allen and Carne
[26].
In what concerns the SWAT method, the generalization presented by Genaro and Rade [20] to
systems with multiple forces acting simultaneously, would be an important step in the improvement
reliability of this method.
It would be also interesting in the context of rotor transient operation to employ these methods
on the detection of changes in the rotor dynamic operation, i.e. to detect changes in the rotor’s spin
speed or even in the spin speed change rate ̈ (see Appendix 6).
One should recall finally that these methods were applied exclusively numerically.
Experimental application would imply necessarily adjustments in the followed procedure, namely in the
definition of the state-space system used in the ISF and DMISF methods. In the mentioned methods, it
would be more advantageous to use the state-space system based on modal properties (easily
accessed in experimental test mounts) rather than based on the system matrices (that implies the
usage of a numeric finite element model).
83
REFERENCES
[1] F. Nelson, Rotor Dynamics without Equations, vol. 10(3), 2007, pp. 2-10.
[2] D. J. Ewins, Control of vibration and resonance in aero engines and rotating machinery - An
overview, vol. 87, Imperial College London, United Kingdom: International Journal of Pressure
Vessels and Piping, 2010, pp. 504-510.
[4] M. Lalanne and G. Ferraris, Rotordynamics prediction in Engineering, Wiley; 2 edition, 1998.
[5] H. P. M. Ernst-Heinrich Hirschel, Aeronautical research in Germany: from Lilienthal until today,
vol. Volume 147, Springer, 2004.
[6] G. Ferraris, V. Maisonneuve and M. Lalanne, Prediction of the Dynamic Behavior of Non-
symmetric Coaxial Co- or Counter-Rotating Rotors, vol. 195(4), Journal of Sound and Vibration,
1996, pp. 649-666.
[7] M. Adams, Rotating Machinery Vibration, 2nd Edition ed., CRC Press, 2009.
[9] E. Assis and V. Steffen Jr., Inverse Problem Techniques For The Identification Of Rotor-Bearing
Systems, vol. 11, Inverse Problems in Engineering, 2003, pp. 39-53.
[10] Q. Han, H. Yao and B. Wen, Parameter identifications for a rotor system based on its finite
element model and with varying speeds, vol. 26, Acta Mech Sin, 2010, pp. 299-303.
[11] K. Stevens, Force Identification Problems - An Overview, Proceedings of the 1987 SEM Spring
Conference on Experimental Mechanics, 1987, pp. 838-844.
[12] M. S. Allen and T. G. Carne, Delayed, multi-step inverse structural filter for robust force
identification, vol. 22, Mechanical Systems and Signal Processing, 2008, pp. 1036-1054.
[14] M. M. Neves and N. M. M. Maia, Estimation of Applied Forces Using The Transmissibility
Concept, vol. paper 377, IDMEC-IST, Technical University of Lisbon, Department of Mechanical
Engineering: ISMA 2010.
[15] Y. Lage, N. M. M. Maia, M. M. Neves and A. M. R. Ribeiro, A Force Identification Approach with
the Transmissibility Concept for Multi-Degree-Of-Freedom Systems, IDMEC, Department of
Mechanical Engineering, Instituto Superior Técnico.
84
[16] N. Maia, M. Fontul and A. Ribeiro, Transmissibility of Forces in Multiple-Degree-of-Freedom
Systems, Leuven, Belgium: Proceedings of ISMA 2006, "Noise and Vibration Engineering", 2006.
[18] T. G. Carne, V. I. Bateman and R. L. Mayes, Force Reconstruction using the sum of weighted
accelerations technique, San Diego, CA: 10th International Modal Analysis Conference (IMAC X),
1992, pp. 291-298.
[20] G. Genaro and D. A. Rade, Input force identification in the time-domain, Santa Barbara, CA: 16th
International Modal Analysis Conference (IMAC XVI), 1998, pp. 124-129.
[21] A. D. Kammer and D. C. Steltzner, Input Force Estimation Using an Inverse Structural Filter, 17th
International Modal Analysis Conference (IMAC XVII), 1999, pp. 954-960.
[22] J. Unger and G. De Roeck, System Identification and Damage Detection in Civil Engineering,
unpublished, 2002.
[23] Z. S. Zutavern and D. W. Childs, Identification of Rotordynamic Forces in a Flexible Rotor System
Using Magnetic Bearings, vol. Vol. 130, Transactions of the ASME, 2008, pp. 022504/1-6.
[24] J. Verhoeven, Excitation force identification of rotating machines using operational rotor/stator
amplitude data and analytical synthesized tranfer functions, vol. 110, Boston: American society of
mechanical engineers, Design engineering technology conferences, 1987, pp. 307-314.
[25] M. Spirig and T. Staubli, Identification of Non-linear Rotor Dynamic Coefficients Using Multiple
Circunfrencial Pressure Measurements, Hochschule Luzern, 1997.
[26] M. S. Allen and T. G. Carne, Comparison of Inverse Structural Filter and Sum of Weighted
Accelerations Technique Time Domain Force Identification Methods, 47th AIAA-ASME-ASCE-
AHS-ASC Structures, Strustural Dynamics and Materials Conference, 2006.
[28] Release Documentation for ANSYS, Theory Reference, Chapter 17. Analysis Procedures, 17.2.
Transient Analysis.
[29] G. H. Hegen, W. Puffert-Meissner, L. Dieterle and H. Vollmers, Flow field investigations on a wing
installed Counter Rotating Ultra-high-bypass Fan engine simulator in the low speed wind tunnel
DNW-LST, Vols. NLR-TP-99104, National Aerospace Laborator y NLR, 1999.
85
APPENDIXES
The equations that depict the orbit motion of the first vibration of a undamped asymmetric rotor
are the expressions (2.1.3.9-10). If these equations are introduced on the expression (2.1.4.3), it
gives:
From the inspection of the equations (2.1.3.9-10), it is visible that the sign of the product
is dependent of the sign of the following expression :
( ( ) )( ( ) ) (A.1.2)
The points of sense of whirl change, are the point at which is null. Therefore the rotating
speeds at which this happens are easily obtained from (A.1.2), giving:
√ (A.1.3)
rotor in FW
rotor in BW
Another interesting aspect, is the location of the critical speeds relative to the whirl change
speeds and . As it was referred in the previous section, the critical speeds are determined
making the denominator of the expressions (2.1.3.9-10) equal to zero. Concretizing this statement, it
results:
( ) ( )( ) ( ) (A.1.4)
( ) ( ) ( ( ) ) (A.1.5)
86
Because function ( ) is a quadratic one we can conclude that it is positive if and
, being and the two critical speeds.
It can be shown that introducing the expressions in (A.1.3) in equation (A.1.4) it results:
( )( )
( )
( )( )
( ) (A.1.6)
So, if the asymmetry in the bearings is so that in both situations of signal (see
the definitions of in equations 2.1.2.15-16), the relations between critical speeds and whirl sense
change speeds are:
with
with
Additionally is curious to verify that the elliptical orbit become progressively flatter on the
axis and more elongated on the axis as the rotor passes the BW region. These situations are all
depicted on the Figures A.1-2.
𝛺 𝜔 𝛺 Ω
𝜔
Figure A.1 – Whirl sense and orbit phase with
87
𝛺 𝜔 𝛺 Ω
𝜔
Figure A.2 – Whirl sense and orbit phase with
[ ]{ ̈ } [ ]{ ̇ } [ ]{ } { ( )} (A.2.1)
whose meaning of the matrices and vectors are already known from previous sections.
To start with, the matrices of equation (A.2.1) are decomposed into their symmetric and skew-
symmetric parts. This is done so that the dynamical effects could be identified as energy conservative
and energy non-conservative.
From the matrix algebra theory, the decomposition of a generic matrix [ ], can be expressed
as follows:
88
[ ] [ ] [ ] (A.2.2)
where:
[ ] [ ]
[ ] [ ] (A.2.3)
The and indexes designate the symmetric and skew-symmetric parts respectively.
In the same way the matrices of (2.1.5.1) are decomposed, resulting:
[ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ]
(A.2.4)
That [ ] and [ ] are conservative, and [ ] non-conservative, can be an accepted
statement [7]. The skew-symmetric parts are now further analyzed, regarding their influence in the
incremental work done on the rotor, so that they could be classified into causers of conservative or
non-conservative energetic effects.
Firstly [ ] will be analyzed. In equation (2.1.2.26), it is visible that the contributions to the
skew-symmetric part of the damping matrix come from two different sources. The first contribution is
always present in the considered model and that is the damping effect resultant from the gyroscopic
effect, the second contribution depends on the skew-symmetric characteristics of the bearings
( and ), coherent with the definition shown in Figure 2.4. The computation of the incremental
work done by these two contributions must be effectuated, so that the energetic effects can be
determined. The incremental work done on the rotor by the skew-symmetric terms of a bearing is
written in the following way:
̇
[ ]{ }{ } (A.2.5)
̇
in which:
[ ] [ ] (A.2.6)
( ̇ ̇ ̇ ̇) (A.2.7)
The incremental work is null, since the force vector is always perpendicular to its associated
velocity vector. Let us now focus on the gyroscopic moment effects in what concerns the incremental
work done on the rotor. The expression that shows this relation is:
̇
[ ]{ }{ }
̇
̇
⏟ [ ]{ }{ ̇ ̇}
̇
(A.2.8)
Again the resulting incremental work is null, because the gyroscopic moment vector is always
perpendicular to its associated angular velocity vector. The conclusion to be taken from (A.2.7) and
(A.2.8), is that the skew-symmetric part of the damping matrix includes only conservative force fields
and doesn’t contribute therefore to the addition or dissipation of energy of the rotor.
Considering again the course of reasoning used to obtain the expression (2.1.2.26), it is visible
that the strain energy is only responsible for the symmetric terms of the stiffness matrix [ ] . So, the
89
skew-symmetric terms [ ] can only be influenced by the skew-symmetric stiffness terms of the
rotor’s bearings.
Again, to determine the energy effect of these terms the calculation of the incremental work is
indispensable:
[ ]{ }{ }
(A.2.9)
Where:
and
Hence, the incremental work done by the skew-symmetric part of the bearings is an exact
differential, meaning that the transferred energy between any two given points of the rotor’s trajectory
is path dependent and therefore, non-conservative.
𝑧 (𝑥)
𝑧 (𝑥)
FigureA.3 – Generic periodic orbit of a rotor relative to a inertial reference frame– source: [7] - edited
Now it must be determined if this non-conservative effect adds or dissipates energy from the
rotor. That can be achieved formulating the expression an expression of the net energy-per-cycle
exchange resultant from the [ ] terms:
∮ ∮( )
(A.2.10)
With the help of the Figure A.3, and considering two line integrals between the generic points
and , the previous integral becomes:
∫ ( )
(A.2.11)
Watching carefully the expression (A.2.11), it is perceivable that the integral represents the
orbit area, and that its value is positive when in co-rotational whirl, and negative otherwise. It should
be noted additionally, that when there’s dissipation of energy , and when there’s energy being
added to the rotor . Since , it can be concluded that the skew-symmetric terms of the
stiffness have a dissipative effect in backward whirl conditions and excites (energy addition) the rotor if
the whirl is forwards.
90
The effects of [ ] terms are now investigated. This terms can only result from the rotor’s
elements mass matrices, namely from those of the disk and the shaft. By similarity with the formulation
followed in equations (A.2.9), (A.2.10) and (A.2.11), it can be obtained for the energy-per-cycle
influence of the skew symmetric part of the mass energy the expression right below:
∮( ̈ ̈ )
(A.2.12)
If the rotor transversal vibration is considered as harmonic, such is the case of equations
(2.1.2.5) and (2.1.2.6). The equation (A.2.12) can be further simplified:
∮( )
(A.2.13)
The factor comes from the second time derivative of the trigonometric functions present in
(2.1.2.5) and (2.1.2.6). Similarly with what was done to obtain equation (A.2.11), the integral in (A.2.13
can be split in two line integrals giving:
∫ ( )
(A.2.14)
For , and recalling the considerations previously written regarding the nature and
signal of the line integral displayed in (A.2.14), one concludes that the skew-symmetric part of the
mass matrix dissipates energy ( ) when the rotor is in forward whirl, and an instability causer
( ) when the opposite is verified. Although this phenomena is theoretically possible, it has
never been documented [7](page 76), meaning that is consistent with physical reality. This
conclusion is significant, since it implies that the rotor’s mass matrix must be constrained to symmetry
in order to reproduce real results.
Neglecting therefore let us now write an expression that includes non-conservative
contributions of the skew-symmetric stiffness matrix and the symmetric damping matrix, in one single
expression, so that the total energy-per-cycle exchange of the rotor can be obtained. It is worth to
mention, that both non-conservative contributions depend only on the bearing stiffness and damping
characteristics.
For a better visualization of the mentioned effects of the symmetric and skew-symmetric parts
Figure A.3, shows two typical rotor orbits, both in FW (a) and (c) and BW (b) and (d).
91
Figure A.4 – Force components of the various matrices decomposed on its symmetric and skew-symmetric terms
– source: [7]
On the circular orbits in Figure A.4 (a) and (b), it is visible that the only force components that
change their sense when the rotor changes from FW to BW, are the vectors ̇ and ̇ . The first
acts always parallel, but in the opposite sense to the instantaneous velocity vector, thus contributing
for the stable behavior of the rotor. The second always acts radially to the rotor’s orbit but has a
centripetal sense if the rotor is in FW and centrifugal sense when the rotor is in BW. This happens
because ̇ is the gyroscopic effect term, and is this effect the responsible for the divergence of the
FW and BW branches of the Campbell Diagram as Figure 2.8 states. The frequency grows with the
rotating speed if the Monorotor is in FW because the gyroscopic effect force vector ̇ has the same
sense of the rotor stiffness force vector , causing virtual structure stiffening and therefore a natural
frequency growth. The opposite happens when the Monorotor is in BW.
It is also observable that the instability generator terms ( ) and ( ̈ ), are acting parallel
and in the sense of motion of the orbit, when it is in FW and BW respectively. This confirms what was
said concerning the non-conservative effects of these terms.
On the same Figure A.4 are represented in (c) and (d) an elliptical rotor orbit shape. The
general conclusions taken before to the circular shape orbits, regarding the sense of action and the
stability effects of each term, can be taken again into account for these orbits, excepting that this time
the forces acting on the rotor are harder to visualize, since the displacement vector is no longer
perpendicular to the instantaneous velocity vector ̇ , with the exception, of course, of the points where
vector overlap the semi-major and semi-minor axis of the ellipse. Consequently, force vector ( )
stops acting exclusively perpendicularly to the instantaneous velocity vector. This fact is particularly
relevant since it explains why, when the orbits are elliptical, and hence when the rotor is asymmetric, a
co-rotational excitation force such as that caused by a mass unbalance present in a rotor’s disk, can
induce both FW and BW vibrations, while it induces in a symmetric rotor (circular orbit) only FW
vibrations.
Let us recall the expression (2.1.2.19), which describes the radial interaction force vector
{ } on a bearing of the rotor. The incremental work done by the forces acting on the bearing is
simply calculated multiplying each component of this force vector by the respective incremental
displacements. The energy-per-cycle exchange on the bearings can then be written as:
92
⁄
∮( ) ⏟ ∫ ( ̇ ̇ )
(A.2.15)
Let us consider a generic harmonic rotor orbit, so that an expression for the energy exchange
per harmonic cycle can be obtained. For this purpose, and will be defined as:
( )
( )
(A.2.16)
Substituting expressions, (A.2.16) and (2.1.2.19) in (A.2.15), solving the resultant integral, and
recalling that the incremental work done by the symmetric part of the stiffness matrix and the skew-
symmetric part of the damping matrix is null, it yields:
[ ( ) ( )]
(A.2.17)
The parameters , , , , and are all positive, furthermore the phase angles
difference is positive for forward whirl ( ( ) ), and negative for backward whirl rotor motion
( ( ) ). Typically the dependent term of the equation (A.2.17), grows faster than the
dependent term, with increasing rotating speed. This means, that for corotational whirl (forward whirl),
the two effects can exactly balance at a given rotating speed . This speed is dubbed the instability
threshold speed, since an increment in the rotating speed would lead the rotor into an energy addition
state or self-excitation.
It becomes therefore clear, from the analysis of the previous equation that unstable
phenomena can only derive from forward whirl state. Moreover, it is worth noting, that instability will
occur first in the lowest-frequency mode, since it is characterized by the largest traversed orbit areas,
and consequently by higher products.
With these equations the fundamentals of energy are concluded. In the first sections of
chapter 2.1 the kinetic and strain energies resultant of the rotor operation were presented, those
equation together with the ones presented just now, that account for the dissipation self-excitation
energy with origin on the bearings, supply the necessary tools for a basic study on rotor’s energy
efficiency.
Instability Critical Speed Determination and Parameters that Affect its Performance
Despite the energy-per-cycle exchange in the bearing capability to detect the stability
threshold speed, it isn’t the usual method used to this end. In fact the eigenvalues of the dynamic
system are the most common tool for instability prediction. This is so because usually in computational
methods there’s an inherent error associated with the extraction of the eigenvectors [7]. This leads to
slightly different orbits and consequently the variables in equations (A.2.16) are affected by this error
contaminating the speed threshold computation through equation (A.2.17).
As it will be shown, the methods based on the determination of the eigenvalues in free motion
conditions are more simple and precise since they don’t involve so much intermediate calculations as
the energy based method of equation (A.2.17).
On a generic damped rotor in free motion, the solutions of the eigenvalue/eigenvector problem
are given as described by the equations (2.1.3.1-2) whose eigenvalues are in the form:
(A.2.18)
√
93
whose variables meanings were already explained.
Introducing equation (A.2.18) in the solution equations (2.1.3.1-2) it results:
( )
√ ( )
( )
√ ( )
(A.2.19)
( )
where the second terms ( ) represent oscillatory motion and the first terms √
, represent
the amplitude decay or increase over time, depending if its exponent is negative or positive
respectively.
The Table A.1 summarizes how will the structural system behave with the change of the
values of and .
Table A.1 – source: [7] - edited
and Signal Modal Motion
Zero damped, steady state sinusoidal motion
Underdamped, sinusoidal, exponential decay
Negatively damped, sinusoidal, exponential growth
Rigid body mode
Overdamped, nonoscillatory, exponential decay
Statically unstable, nonoscillatory, exponential growth
As the table shows, a dynamic system will enter into instability whenever the value of is
negative, since it changes makes the real part of the eigenvalue positive.
So for a simple damped rotor, one can compute the eigenvalues for several rotating speeds
until the real part becomes real. This method besides giving the critical speed for instability would also
give the value of the natural frequency at which an unstable state is verified.
Another method present in the literature [4] is the Routh-Hurwitz criterion. In order to make this
method clear, let us consider again equation (2.1.3.4) for a Multirotor. This equation can be generically
described as the following polynomial equation:
(A.2.20)
(A.2.21)
[ ]
94
[ ] (A.2.22)
from which are deductible the Routh-Hurwitz coefficients for Multirotor system:
[ ]
[ ] | |
[ ]
[ ] (A.2.23)
( ) ( ) ( )
( ) ( )( )
( ),
(A.2.24)
Substituting the expressions of (A.2.24) into the Routh-Hurwitz coefficients whose calculation
is expressed in (A.2.23), it yields:
( )
( )( )
( )
( )( ) ( )
[ ]
(A.2.25)
Routh and Hurwitz have demonstrated that the system is unstable whenever one of the Routh-
Hurwitz coefficients is negative [4].
Note that if the couple terms of the bearings’ stiffness are equal to zero, the Multirotor is stable
for any value of , which confirms the statements made after the deduction of equation (A.2.17).
95
describe the first bending mode of a simple supported beam. For this function the deformed shape of
the rotor would have its maximum value at the middle point of the shaft.
In rotors with complex geometries, where, for instance, the disks and elastic bearings are
deviated from the shaft’s middle point, and an additional outer shaft is considered the first mode shape
can be slightly different from that described by the shape function in equation (2.1.2.1).
Let us watch closely the first bending vibration mode shape of a symmetric Multirotor such as
that used in chapter 4.1.4 at rest. For this purpose we will use a finite element model and the ANSYS®
toolset, and the refined mesh presented on chapter 4.1.2.
The model’s nodes along with the modal shape normalized to the mass matrix are presented
in the Figure A.5. while the total nodal displacement is presented in Table A.2 for each node number.
Outer shaft
In Figure A.5 it is visible that the middle point of the inner shaft corresponds to node 16, but
the displacement Table A.2 shows that the maximum displacement is verified in node 15. This slight
difference in the mode’s shape can contribute decisively for the disparities in the computation of
natural frequencies of the rotor with each of the used methods.
Here it is intended to explain the reason way some additional asymptotes appear in the
Response diagrams where the finite element method was employed and to explain way some critical
96
resonance speeds in the response diagrams are slightly deviated from the predicted critical speeds in
the Campbell diagrams.
This is actually related with the methods used in ANSYS® to determine both the Campbell
and the Response diagrams. As it was said in chapter 4.1 the Campbell diagrams were computed
using a pseudo-modal method and the response diagram was obtained by performing a full harmonic
analysis which computes the structure’s modes through the direct method and considers all the
structure’s modes in its construction.
To show this, the response diagram shown in Figure 4.5 is again considered and reproduced
in Figure A.6.
Figure A.6
Now the resultant Campbell diagram for the correspondent situation calculated with the direct
method instead of the pseudo-modal method is displayed in Figure A.7. The relevant critical speeds
are in the annexed table A.3.
Figure A.7
97
Table A.3
Outer shaft synchronous critical speeds
(rpm)
Direct Method Pseudo-modal Error Direct/Pseudo-modal methods (%)
1-FW 2-BW 2-FW 1-FW 2-BW 1-FW 2-BW
1906 2009 5018 1908 2012 0.105 0.149
As it is visible in Figure A.7, the anti-resonance observed in the Response Diagram in Figure
A.6 is related with the proximity of the critical speed of the second bending vibration mode FW.
Additionally note that the natural frequencies obtained with the direct method have slightly
lower values than that obtained with the pseudo-modal method. This difference is accentuated with
growing rotating speed.
The Figure A.8 shows the three degree-of-freedom per node beam element implemented in
the Matlab® code used on the application chapter 4.2.1. As it is also visible in the mentioned figure,
the beam element has two nodes.
[ ]
(A.5.1)
[ ] (A.5.2)
[ ]
98
⁄ ⁄
[ ] (A.5.3)
⁄ ⁄
[ ]
∫ ( ̇ ̇ ) ∫ ( ̇ ̇ ) ̇ ̇∫ ̇
(A.6.3)
If the Lagrange equation is applied with the considered displacement vector (3.1.1.2) in
chapter 3.1.1, to the disk’s and the shaft’s kinetic energy expressions (A.6.2-3), it results.
( ) ̈ ̇ ̇ ̈
̇
( ̇ ) ( ) ̈ ̇ ̇ ̈ (A.6.4)
where is the mass matrix in equation (3.1.2.1) and the gyroscopic effect matrix also in
(3.1.2.1) but divided by . ( ) correspond to the matrices with the same nomenclature
presented on chapter 3.1.2 and is the matrix (3.1.2.9) divided by .
Matrices and are two additional terms present only in transient motion defined as:
[ ] (A.6.5)
99
(A.6.6)
[ ]
In what concerns the mass unbalance, its kinetic energy expression for transient motions is
given by:
̇( ̇ ̇ ) (A.6.7)
̈ ̇
( ) [ ] (A.6.8)
̇ ̈ ̇
To describe the evolution of the rotating speed with time, a linear law was chosen,
characterized by the following expression:
̇( ) (A.6.9)
in which and are given by:
where , , , are initial conditions of the transient problem and denote respectively the initial and
final rotating speeds and time instants of the transient motion.
Finally it is worth defining the following expressions:
̈ (A.6.10)
( ) (A.6.11)
with:
100